git.oblomov.eu Git - linux-2.6/blob - 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/init.h>
  57 #include <linux/capability.h>
  58 #include <linux/file.h>
  59 #include <linux/string.h>
  60 #include <linux/seq_file.h>
  61 #include <linux/namei.h>
  62 #include <linux/mnt_namespace.h>
  63 #include <linux/mm.h>
  64 #include <linux/rcupdate.h>
  65 #include <linux/kallsyms.h>
  66 #include <linux/resource.h>
  67 #include <linux/module.h>
  68 #include <linux/mount.h>
  69 #include <linux/security.h>
  70 #include <linux/ptrace.h>
  71 #include <linux/cgroup.h>
  72 #include <linux/cpuset.h>
  73 #include <linux/audit.h>
  74 #include <linux/poll.h>
  75 #include <linux/nsproxy.h>
  76 #include <linux/oom.h>
  77 #include <linux/elf.h>
  78 #include <linux/pid_namespace.h>
  79 #include "internal.h"
  80
  81 /* NOTE:
  82  *      Implementing inode permission operations in /proc is almost
  83  *      certainly an error.  Permission checks need to happen during
  84  *      each system call not at open time.  The reason is that most of
  85  *      what we wish to check for permissions in /proc varies at runtime.
  86  *
  87  *      The classic example of a problem is opening file descriptors
  88  *      in /proc for a task before it execs a suid executable.
  89  */
  90
  91 struct pid_entry {
  92         char *name;
  93         int len;
  94         mode_t mode;
  95         const struct inode_operations *iop;
  96         const struct file_operations *fop;
  97         union proc_op op;
  98 };
  99
 100 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 101         .name = (NAME),                                 \
 102         .len  = sizeof(NAME) - 1,                       \
 103         .mode = MODE,                                   \
 104         .iop  = IOP,                                    \
 105         .fop  = FOP,                                    \
 106         .op   = OP,                                     \
 107 }
 108
 109 #define DIR(NAME, MODE, OTYPE)                                                  \
 110         NOD(NAME, (S_IFDIR|(MODE)),                                             \
 111                 &proc_##OTYPE##_inode_operations, &proc_##OTYPE##_operations,   \
 112                 {} )
 113 #define LNK(NAME, OTYPE)                                        \
 114         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 115                 &proc_pid_link_inode_operations, NULL,          \
 116                 { .proc_get_link = &proc_##OTYPE##_link } )
 117 #define REG(NAME, MODE, OTYPE)                          \
 118         NOD(NAME, (S_IFREG|(MODE)), NULL,               \
 119                 &proc_##OTYPE##_operations, {})
 120 #define INF(NAME, MODE, OTYPE)                          \
 121         NOD(NAME, (S_IFREG|(MODE)),                     \
 122                 NULL, &proc_info_file_operations,       \
 123                 { .proc_read = &proc_##OTYPE } )
 124 #define ONE(NAME, MODE, OTYPE)                          \
 125         NOD(NAME, (S_IFREG|(MODE)),                     \
 126                 NULL, &proc_single_file_operations,     \
 127                 { .proc_show = &proc_##OTYPE } )
 128
 129 int maps_protect;
 130 EXPORT_SYMBOL(maps_protect);
 131
 132 static struct fs_struct *get_fs_struct(struct task_struct *task)
 133 {
 134         struct fs_struct *fs;
 135         task_lock(task);
 136         fs = task->fs;
 137         if(fs)
 138                 atomic_inc(&fs->count);
 139         task_unlock(task);
 140         return fs;
 141 }
 142
 143 static int get_nr_threads(struct task_struct *tsk)
 144 {
 145         /* Must be called with the rcu_read_lock held */
 146         unsigned long flags;
 147         int count = 0;
 148
 149         if (lock_task_sighand(tsk, &flags)) {
 150                 count = atomic_read(&tsk->signal->count);
 151                 unlock_task_sighand(tsk, &flags);
 152         }
 153         return count;
 154 }
 155
 156 static int proc_cwd_link(struct inode *inode, struct path *path)
 157 {
 158         struct task_struct *task = get_proc_task(inode);
 159         struct fs_struct *fs = NULL;
 160         int result = -ENOENT;
 161
 162         if (task) {
 163                 fs = get_fs_struct(task);
 164                 put_task_struct(task);
 165         }
 166         if (fs) {
 167                 read_lock(&fs->lock);
 168                 *path = fs->pwd;
 169                 path_get(&fs->pwd);
 170                 read_unlock(&fs->lock);
 171                 result = 0;
 172                 put_fs_struct(fs);
 173         }
 174         return result;
 175 }
 176
 177 static int proc_root_link(struct inode *inode, struct path *path)
 178 {
 179         struct task_struct *task = get_proc_task(inode);
 180         struct fs_struct *fs = NULL;
 181         int result = -ENOENT;
 182
 183         if (task) {
 184                 fs = get_fs_struct(task);
 185                 put_task_struct(task);
 186         }
 187         if (fs) {
 188                 read_lock(&fs->lock);
 189                 *path = fs->root;
 190                 path_get(&fs->root);
 191                 read_unlock(&fs->lock);
 192                 result = 0;
 193                 put_fs_struct(fs);
 194         }
 195         return result;
 196 }
 197
 198 #define MAY_PTRACE(task) \
 199         (task == current || \
 200         (task->parent == current && \
 201         (task->ptrace & PT_PTRACED) && \
 202          (task_is_stopped_or_traced(task)) && \
 203          security_ptrace(current,task) == 0))
 204
 205 struct mm_struct *mm_for_maps(struct task_struct *task)
 206 {
 207         struct mm_struct *mm = get_task_mm(task);
 208         if (!mm)
 209                 return NULL;
 210         down_read(&mm->mmap_sem);
 211         task_lock(task);
 212         if (task->mm != mm)
 213                 goto out;
 214         if (task->mm != current->mm && __ptrace_may_attach(task) < 0)
 215                 goto out;
 216         task_unlock(task);
 217         return mm;
 218 out:
 219         task_unlock(task);
 220         up_read(&mm->mmap_sem);
 221         mmput(mm);
 222         return NULL;
 223 }
 224
 225 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 226 {
 227         int res = 0;
 228         unsigned int len;
 229         struct mm_struct *mm = get_task_mm(task);
 230         if (!mm)
 231                 goto out;
 232         if (!mm->arg_end)
 233                 goto out_mm;    /* Shh! No looking before we're done */
 234
 235         len = mm->arg_end - mm->arg_start;
 236
 237         if (len > PAGE_SIZE)
 238                 len = PAGE_SIZE;
 239
 240         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 241
 242         // If the nul at the end of args has been overwritten, then
 243         // assume application is using setproctitle(3).
 244         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 245                 len = strnlen(buffer, res);
 246                 if (len < res) {
 247                     res = len;
 248                 } else {
 249                         len = mm->env_end - mm->env_start;
 250                         if (len > PAGE_SIZE - res)
 251                                 len = PAGE_SIZE - res;
 252                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 253                         res = strnlen(buffer, res);
 254                 }
 255         }
 256 out_mm:
 257         mmput(mm);
 258 out:
 259         return res;
 260 }
 261
 262 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 263 {
 264         int res = 0;
 265         struct mm_struct *mm = get_task_mm(task);
 266         if (mm) {
 267                 unsigned int nwords = 0;
 268                 do
 269                         nwords += 2;
 270                 while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 271                 res = nwords * sizeof(mm->saved_auxv[0]);
 272                 if (res > PAGE_SIZE)
 273                         res = PAGE_SIZE;
 274                 memcpy(buffer, mm->saved_auxv, res);
 275                 mmput(mm);
 276         }
 277         return res;
 278 }
 279
 280
 281 #ifdef CONFIG_KALLSYMS
 282 /*
 283  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 284  * Returns the resolved symbol.  If that fails, simply return the address.
 285  */
 286 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 287 {
 288         unsigned long wchan;
 289         char symname[KSYM_NAME_LEN];
 290
 291         wchan = get_wchan(task);
 292
 293         if (lookup_symbol_name(wchan, symname) < 0)
 294                 return sprintf(buffer, "%lu", wchan);
 295         else
 296                 return sprintf(buffer, "%s", symname);
 297 }
 298 #endif /* CONFIG_KALLSYMS */
 299
 300 #ifdef CONFIG_SCHEDSTATS
 301 /*
 302  * Provides /proc/PID/schedstat
 303  */
 304 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 305 {
 306         return sprintf(buffer, "%llu %llu %lu\n",
 307                         task->sched_info.cpu_time,
 308                         task->sched_info.run_delay,
 309                         task->sched_info.pcount);
 310 }
 311 #endif
 312
 313 #ifdef CONFIG_LATENCYTOP
 314 static int lstats_show_proc(struct seq_file *m, void *v)
 315 {
 316         int i;
 317         struct task_struct *task = m->private;
 318         seq_puts(m, "Latency Top version : v0.1\n");
 319
 320         for (i = 0; i < 32; i++) {
 321                 if (task->latency_record[i].backtrace[0]) {
 322                         int q;
 323                         seq_printf(m, "%i %li %li ",
 324                                 task->latency_record[i].count,
 325                                 task->latency_record[i].time,
 326                                 task->latency_record[i].max);
 327                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 328                                 char sym[KSYM_NAME_LEN];
 329                                 char *c;
 330                                 if (!task->latency_record[i].backtrace[q])
 331                                         break;
 332                                 if (task->latency_record[i].backtrace[q] == ULONG_MAX)
 333                                         break;
 334                                 sprint_symbol(sym, task->latency_record[i].backtrace[q]);
 335                                 c = strchr(sym, '+');
 336                                 if (c)
 337                                         *c = 0;
 338                                 seq_printf(m, "%s ", sym);
 339                         }
 340                         seq_printf(m, "\n");
 341                 }
 342
 343         }
 344         return 0;
 345 }
 346
 347 static int lstats_open(struct inode *inode, struct file *file)
 348 {
 349         int ret;
 350         struct seq_file *m;
 351         struct task_struct *task = get_proc_task(inode);
 352
 353         ret = single_open(file, lstats_show_proc, NULL);
 354         if (!ret) {
 355                 m = file->private_data;
 356                 m->private = task;
 357         }
 358         return ret;
 359 }
 360
 361 static ssize_t lstats_write(struct file *file, const char __user *buf,
 362                             size_t count, loff_t *offs)
 363 {
 364         struct seq_file *m;
 365         struct task_struct *task;
 366
 367         m = file->private_data;
 368         task = m->private;
 369         clear_all_latency_tracing(task);
 370
 371         return count;
 372 }
 373
 374 static const struct file_operations proc_lstats_operations = {
 375         .open           = lstats_open,
 376         .read           = seq_read,
 377         .write          = lstats_write,
 378         .llseek         = seq_lseek,
 379         .release        = single_release,
 380 };
 381
 382 #endif
 383
 384 /* The badness from the OOM killer */
 385 unsigned long badness(struct task_struct *p, unsigned long uptime);
 386 static int proc_oom_score(struct task_struct *task, char *buffer)
 387 {
 388         unsigned long points;
 389         struct timespec uptime;
 390
 391         do_posix_clock_monotonic_gettime(&uptime);
 392         read_lock(&tasklist_lock);
 393         points = badness(task, uptime.tv_sec);
 394         read_unlock(&tasklist_lock);
 395         return sprintf(buffer, "%lu\n", points);
 396 }
 397
 398 struct limit_names {
 399         char *name;
 400         char *unit;
 401 };
 402
 403 static const struct limit_names lnames[RLIM_NLIMITS] = {
 404         [RLIMIT_CPU] = {"Max cpu time", "ms"},
 405         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 406         [RLIMIT_DATA] = {"Max data size", "bytes"},
 407         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 408         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 409         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 410         [RLIMIT_NPROC] = {"Max processes", "processes"},
 411         [RLIMIT_NOFILE] = {"Max open files", "files"},
 412         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 413         [RLIMIT_AS] = {"Max address space", "bytes"},
 414         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 415         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 416         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 417         [RLIMIT_NICE] = {"Max nice priority", NULL},
 418         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 419 };
 420
 421 /* Display limits for a process */
 422 static int proc_pid_limits(struct task_struct *task, char *buffer)
 423 {
 424         unsigned int i;
 425         int count = 0;
 426         unsigned long flags;
 427         char *bufptr = buffer;
 428
 429         struct rlimit rlim[RLIM_NLIMITS];
 430
 431         rcu_read_lock();
 432         if (!lock_task_sighand(task,&flags)) {
 433                 rcu_read_unlock();
 434                 return 0;
 435         }
 436         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 437         unlock_task_sighand(task, &flags);
 438         rcu_read_unlock();
 439
 440         /*
 441          * print the file header
 442          */
 443         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 444                         "Limit", "Soft Limit", "Hard Limit", "Units");
 445
 446         for (i = 0; i < RLIM_NLIMITS; i++) {
 447                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 448                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 449                                          lnames[i].name, "unlimited");
 450                 else
 451                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 452                                          lnames[i].name, rlim[i].rlim_cur);
 453
 454                 if (rlim[i].rlim_max == RLIM_INFINITY)
 455                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 456                 else
 457                         count += sprintf(&bufptr[count], "%-20lu ",
 458                                          rlim[i].rlim_max);
 459
 460                 if (lnames[i].unit)
 461                         count += sprintf(&bufptr[count], "%-10s\n",
 462                                          lnames[i].unit);
 463                 else
 464                         count += sprintf(&bufptr[count], "\n");
 465         }
 466
 467         return count;
 468 }
 469
 470 /************************************************************************/
 471 /*                       Here the fs part begins                        */
 472 /************************************************************************/
 473
 474 /* permission checks */
 475 static int proc_fd_access_allowed(struct inode *inode)
 476 {
 477         struct task_struct *task;
 478         int allowed = 0;
 479         /* Allow access to a task's file descriptors if it is us or we
 480          * may use ptrace attach to the process and find out that
 481          * information.
 482          */
 483         task = get_proc_task(inode);
 484         if (task) {
 485                 allowed = ptrace_may_attach(task);
 486                 put_task_struct(task);
 487         }
 488         return allowed;
 489 }
 490
 491 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
 492 {
 493         int error;
 494         struct inode *inode = dentry->d_inode;
 495
 496         if (attr->ia_valid & ATTR_MODE)
 497                 return -EPERM;
 498
 499         error = inode_change_ok(inode, attr);
 500         if (!error)
 501                 error = inode_setattr(inode, attr);
 502         return error;
 503 }
 504
 505 static const struct inode_operations proc_def_inode_operations = {
 506         .setattr        = proc_setattr,
 507 };
 508
 509 extern const struct seq_operations mounts_op;
 510 struct proc_mounts {
 511         struct seq_file m;
 512         int event;
 513 };
 514
 515 static int mounts_open(struct inode *inode, struct file *file)
 516 {
 517         struct task_struct *task = get_proc_task(inode);
 518         struct nsproxy *nsp;
 519         struct mnt_namespace *ns = NULL;
 520         struct proc_mounts *p;
 521         int ret = -EINVAL;
 522
 523         if (task) {
 524                 rcu_read_lock();
 525                 nsp = task_nsproxy(task);
 526                 if (nsp) {
 527                         ns = nsp->mnt_ns;
 528                         if (ns)
 529                                 get_mnt_ns(ns);
 530                 }
 531                 rcu_read_unlock();
 532
 533                 put_task_struct(task);
 534         }
 535
 536         if (ns) {
 537                 ret = -ENOMEM;
 538                 p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 539                 if (p) {
 540                         file->private_data = &p->m;
 541                         ret = seq_open(file, &mounts_op);
 542                         if (!ret) {
 543                                 p->m.private = ns;
 544                                 p->event = ns->event;
 545                                 return 0;
 546                         }
 547                         kfree(p);
 548                 }
 549                 put_mnt_ns(ns);
 550         }
 551         return ret;
 552 }
 553
 554 static int mounts_release(struct inode *inode, struct file *file)
 555 {
 556         struct seq_file *m = file->private_data;
 557         struct mnt_namespace *ns = m->private;
 558         put_mnt_ns(ns);
 559         return seq_release(inode, file);
 560 }
 561
 562 static unsigned mounts_poll(struct file *file, poll_table *wait)
 563 {
 564         struct proc_mounts *p = file->private_data;
 565         struct mnt_namespace *ns = p->m.private;
 566         unsigned res = 0;
 567
 568         poll_wait(file, &ns->poll, wait);
 569
 570         spin_lock(&vfsmount_lock);
 571         if (p->event != ns->event) {
 572                 p->event = ns->event;
 573                 res = POLLERR;
 574         }
 575         spin_unlock(&vfsmount_lock);
 576
 577         return res;
 578 }
 579
 580 static const struct file_operations proc_mounts_operations = {
 581         .open           = mounts_open,
 582         .read           = seq_read,
 583         .llseek         = seq_lseek,
 584         .release        = mounts_release,
 585         .poll           = mounts_poll,
 586 };
 587
 588 extern const struct seq_operations mountstats_op;
 589 static int mountstats_open(struct inode *inode, struct file *file)
 590 {
 591         int ret = seq_open(file, &mountstats_op);
 592
 593         if (!ret) {
 594                 struct seq_file *m = file->private_data;
 595                 struct nsproxy *nsp;
 596                 struct mnt_namespace *mnt_ns = NULL;
 597                 struct task_struct *task = get_proc_task(inode);
 598
 599                 if (task) {
 600                         rcu_read_lock();
 601                         nsp = task_nsproxy(task);
 602                         if (nsp) {
 603                                 mnt_ns = nsp->mnt_ns;
 604                                 if (mnt_ns)
 605                                         get_mnt_ns(mnt_ns);
 606                         }
 607                         rcu_read_unlock();
 608
 609                         put_task_struct(task);
 610                 }
 611
 612                 if (mnt_ns)
 613                         m->private = mnt_ns;
 614                 else {
 615                         seq_release(inode, file);
 616                         ret = -EINVAL;
 617                 }
 618         }
 619         return ret;
 620 }
 621
 622 static const struct file_operations proc_mountstats_operations = {
 623         .open           = mountstats_open,
 624         .read           = seq_read,
 625         .llseek         = seq_lseek,
 626         .release        = mounts_release,
 627 };
 628
 629 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 630
 631 static ssize_t proc_info_read(struct file * file, char __user * buf,
 632                           size_t count, loff_t *ppos)
 633 {
 634         struct inode * inode = file->f_path.dentry->d_inode;
 635         unsigned long page;
 636         ssize_t length;
 637         struct task_struct *task = get_proc_task(inode);
 638
 639         length = -ESRCH;
 640         if (!task)
 641                 goto out_no_task;
 642
 643         if (count > PROC_BLOCK_SIZE)
 644                 count = PROC_BLOCK_SIZE;
 645
 646         length = -ENOMEM;
 647         if (!(page = __get_free_page(GFP_TEMPORARY)))
 648                 goto out;
 649
 650         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 651
 652         if (length >= 0)
 653                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 654         free_page(page);
 655 out:
 656         put_task_struct(task);
 657 out_no_task:
 658         return length;
 659 }
 660
 661 static const struct file_operations proc_info_file_operations = {
 662         .read           = proc_info_read,
 663 };
 664
 665 static int proc_single_show(struct seq_file *m, void *v)
 666 {
 667         struct inode *inode = m->private;
 668         struct pid_namespace *ns;
 669         struct pid *pid;
 670         struct task_struct *task;
 671         int ret;
 672
 673         ns = inode->i_sb->s_fs_info;
 674         pid = proc_pid(inode);
 675         task = get_pid_task(pid, PIDTYPE_PID);
 676         if (!task)
 677                 return -ESRCH;
 678
 679         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 680
 681         put_task_struct(task);
 682         return ret;
 683 }
 684
 685 static int proc_single_open(struct inode *inode, struct file *filp)
 686 {
 687         int ret;
 688         ret = single_open(filp, proc_single_show, NULL);
 689         if (!ret) {
 690                 struct seq_file *m = filp->private_data;
 691
 692                 m->private = inode;
 693         }
 694         return ret;
 695 }
 696
 697 static const struct file_operations proc_single_file_operations = {
 698         .open           = proc_single_open,
 699         .read           = seq_read,
 700         .llseek         = seq_lseek,
 701         .release        = single_release,
 702 };
 703
 704 static int mem_open(struct inode* inode, struct file* file)
 705 {
 706         file->private_data = (void*)((long)current->self_exec_id);
 707         return 0;
 708 }
 709
 710 static ssize_t mem_read(struct file * file, char __user * buf,
 711                         size_t count, loff_t *ppos)
 712 {
 713         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 714         char *page;
 715         unsigned long src = *ppos;
 716         int ret = -ESRCH;
 717         struct mm_struct *mm;
 718
 719         if (!task)
 720                 goto out_no_task;
 721
 722         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 723                 goto out;
 724
 725         ret = -ENOMEM;
 726         page = (char *)__get_free_page(GFP_TEMPORARY);
 727         if (!page)
 728                 goto out;
 729
 730         ret = 0;
 731
 732         mm = get_task_mm(task);
 733         if (!mm)
 734                 goto out_free;
 735
 736         ret = -EIO;
 737
 738         if (file->private_data != (void*)((long)current->self_exec_id))
 739                 goto out_put;
 740
 741         ret = 0;
 742
 743         while (count > 0) {
 744                 int this_len, retval;
 745
 746                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 747                 retval = access_process_vm(task, src, page, this_len, 0);
 748                 if (!retval || !MAY_PTRACE(task) || !ptrace_may_attach(task)) {
 749                         if (!ret)
 750                                 ret = -EIO;
 751                         break;
 752                 }
 753
 754                 if (copy_to_user(buf, page, retval)) {
 755                         ret = -EFAULT;
 756                         break;
 757                 }
 758
 759                 ret += retval;
 760                 src += retval;
 761                 buf += retval;
 762                 count -= retval;
 763         }
 764         *ppos = src;
 765
 766 out_put:
 767         mmput(mm);
 768 out_free:
 769         free_page((unsigned long) page);
 770 out:
 771         put_task_struct(task);
 772 out_no_task:
 773         return ret;
 774 }
 775
 776 #define mem_write NULL
 777
 778 #ifndef mem_write
 779 /* This is a security hazard */
 780 static ssize_t mem_write(struct file * file, const char __user *buf,
 781                          size_t count, loff_t *ppos)
 782 {
 783         int copied;
 784         char *page;
 785         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 786         unsigned long dst = *ppos;
 787
 788         copied = -ESRCH;
 789         if (!task)
 790                 goto out_no_task;
 791
 792         if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
 793                 goto out;
 794
 795         copied = -ENOMEM;
 796         page = (char *)__get_free_page(GFP_TEMPORARY);
 797         if (!page)
 798                 goto out;
 799
 800         copied = 0;
 801         while (count > 0) {
 802                 int this_len, retval;
 803
 804                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 805                 if (copy_from_user(page, buf, this_len)) {
 806                         copied = -EFAULT;
 807                         break;
 808                 }
 809                 retval = access_process_vm(task, dst, page, this_len, 1);
 810                 if (!retval) {
 811                         if (!copied)
 812                                 copied = -EIO;
 813                         break;
 814                 }
 815                 copied += retval;
 816                 buf += retval;
 817                 dst += retval;
 818                 count -= retval;
 819         }
 820         *ppos = dst;
 821         free_page((unsigned long) page);
 822 out:
 823         put_task_struct(task);
 824 out_no_task:
 825         return copied;
 826 }
 827 #endif
 828
 829 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 830 {
 831         switch (orig) {
 832         case 0:
 833                 file->f_pos = offset;
 834                 break;
 835         case 1:
 836                 file->f_pos += offset;
 837                 break;
 838         default:
 839                 return -EINVAL;
 840         }
 841         force_successful_syscall_return();
 842         return file->f_pos;
 843 }
 844
 845 static const struct file_operations proc_mem_operations = {
 846         .llseek         = mem_lseek,
 847         .read           = mem_read,
 848         .write          = mem_write,
 849         .open           = mem_open,
 850 };
 851
 852 static ssize_t environ_read(struct file *file, char __user *buf,
 853                         size_t count, loff_t *ppos)
 854 {
 855         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 856         char *page;
 857         unsigned long src = *ppos;
 858         int ret = -ESRCH;
 859         struct mm_struct *mm;
 860
 861         if (!task)
 862                 goto out_no_task;
 863
 864         if (!ptrace_may_attach(task))
 865                 goto out;
 866
 867         ret = -ENOMEM;
 868         page = (char *)__get_free_page(GFP_TEMPORARY);
 869         if (!page)
 870                 goto out;
 871
 872         ret = 0;
 873
 874         mm = get_task_mm(task);
 875         if (!mm)
 876                 goto out_free;
 877
 878         while (count > 0) {
 879                 int this_len, retval, max_len;
 880
 881                 this_len = mm->env_end - (mm->env_start + src);
 882
 883                 if (this_len <= 0)
 884                         break;
 885
 886                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 887                 this_len = (this_len > max_len) ? max_len : this_len;
 888
 889                 retval = access_process_vm(task, (mm->env_start + src),
 890                         page, this_len, 0);
 891
 892                 if (retval <= 0) {
 893                         ret = retval;
 894                         break;
 895                 }
 896
 897                 if (copy_to_user(buf, page, retval)) {
 898                         ret = -EFAULT;
 899                         break;
 900                 }
 901
 902                 ret += retval;
 903                 src += retval;
 904                 buf += retval;
 905                 count -= retval;
 906         }
 907         *ppos = src;
 908
 909         mmput(mm);
 910 out_free:
 911         free_page((unsigned long) page);
 912 out:
 913         put_task_struct(task);
 914 out_no_task:
 915         return ret;
 916 }
 917
 918 static const struct file_operations proc_environ_operations = {
 919         .read           = environ_read,
 920 };
 921
 922 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 923                                 size_t count, loff_t *ppos)
 924 {
 925         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 926         char buffer[PROC_NUMBUF];
 927         size_t len;
 928         int oom_adjust;
 929
 930         if (!task)
 931                 return -ESRCH;
 932         oom_adjust = task->oomkilladj;
 933         put_task_struct(task);
 934
 935         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
 936
 937         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 938 }
 939
 940 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 941                                 size_t count, loff_t *ppos)
 942 {
 943         struct task_struct *task;
 944         char buffer[PROC_NUMBUF], *end;
 945         int oom_adjust;
 946
 947         memset(buffer, 0, sizeof(buffer));
 948         if (count > sizeof(buffer) - 1)
 949                 count = sizeof(buffer) - 1;
 950         if (copy_from_user(buffer, buf, count))
 951                 return -EFAULT;
 952         oom_adjust = simple_strtol(buffer, &end, 0);
 953         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
 954              oom_adjust != OOM_DISABLE)
 955                 return -EINVAL;
 956         if (*end == '\n')
 957                 end++;
 958         task = get_proc_task(file->f_path.dentry->d_inode);
 959         if (!task)
 960                 return -ESRCH;
 961         if (oom_adjust < task->oomkilladj && !capable(CAP_SYS_RESOURCE)) {
 962                 put_task_struct(task);
 963                 return -EACCES;
 964         }
 965         task->oomkilladj = oom_adjust;
 966         put_task_struct(task);
 967         if (end - buffer == 0)
 968                 return -EIO;
 969         return end - buffer;
 970 }
 971
 972 static const struct file_operations proc_oom_adjust_operations = {
 973         .read           = oom_adjust_read,
 974         .write          = oom_adjust_write,
 975 };
 976
 977 #ifdef CONFIG_AUDITSYSCALL
 978 #define TMPBUFLEN 21
 979 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
 980                                   size_t count, loff_t *ppos)
 981 {
 982         struct inode * inode = file->f_path.dentry->d_inode;
 983         struct task_struct *task = get_proc_task(inode);
 984         ssize_t length;
 985         char tmpbuf[TMPBUFLEN];
 986
 987         if (!task)
 988                 return -ESRCH;
 989         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
 990                                 audit_get_loginuid(task));
 991         put_task_struct(task);
 992         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
 993 }
 994
 995 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
 996                                    size_t count, loff_t *ppos)
 997 {
 998         struct inode * inode = file->f_path.dentry->d_inode;
 999         char *page, *tmp;
1000         ssize_t length;
1001         uid_t loginuid;
1002
1003         if (!capable(CAP_AUDIT_CONTROL))
1004                 return -EPERM;
1005
1006         if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
1007                 return -EPERM;
1008
1009         if (count >= PAGE_SIZE)
1010                 count = PAGE_SIZE - 1;
1011
1012         if (*ppos != 0) {
1013                 /* No partial writes. */
1014                 return -EINVAL;
1015         }
1016         page = (char*)__get_free_page(GFP_TEMPORARY);
1017         if (!page)
1018                 return -ENOMEM;
1019         length = -EFAULT;
1020         if (copy_from_user(page, buf, count))
1021                 goto out_free_page;
1022
1023         page[count] = '\0';
1024         loginuid = simple_strtoul(page, &tmp, 10);
1025         if (tmp == page) {
1026                 length = -EINVAL;
1027                 goto out_free_page;
1028
1029         }
1030         length = audit_set_loginuid(current, loginuid);
1031         if (likely(length == 0))
1032                 length = count;
1033
1034 out_free_page:
1035         free_page((unsigned long) page);
1036         return length;
1037 }
1038
1039 static const struct file_operations proc_loginuid_operations = {
1040         .read           = proc_loginuid_read,
1041         .write          = proc_loginuid_write,
1042 };
1043 #endif
1044
1045 #ifdef CONFIG_FAULT_INJECTION
1046 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1047                                       size_t count, loff_t *ppos)
1048 {
1049         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1050         char buffer[PROC_NUMBUF];
1051         size_t len;
1052         int make_it_fail;
1053
1054         if (!task)
1055                 return -ESRCH;
1056         make_it_fail = task->make_it_fail;
1057         put_task_struct(task);
1058
1059         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1060
1061         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1062 }
1063
1064 static ssize_t proc_fault_inject_write(struct file * file,
1065                         const char __user * buf, size_t count, loff_t *ppos)
1066 {
1067         struct task_struct *task;
1068         char buffer[PROC_NUMBUF], *end;
1069         int make_it_fail;
1070
1071         if (!capable(CAP_SYS_RESOURCE))
1072                 return -EPERM;
1073         memset(buffer, 0, sizeof(buffer));
1074         if (count > sizeof(buffer) - 1)
1075                 count = sizeof(buffer) - 1;
1076         if (copy_from_user(buffer, buf, count))
1077                 return -EFAULT;
1078         make_it_fail = simple_strtol(buffer, &end, 0);
1079         if (*end == '\n')
1080                 end++;
1081         task = get_proc_task(file->f_dentry->d_inode);
1082         if (!task)
1083                 return -ESRCH;
1084         task->make_it_fail = make_it_fail;
1085         put_task_struct(task);
1086         if (end - buffer == 0)
1087                 return -EIO;
1088         return end - buffer;
1089 }
1090
1091 static const struct file_operations proc_fault_inject_operations = {
1092         .read           = proc_fault_inject_read,
1093         .write          = proc_fault_inject_write,
1094 };
1095 #endif
1096
1097
1098 #ifdef CONFIG_SCHED_DEBUG
1099 /*
1100  * Print out various scheduling related per-task fields:
1101  */
1102 static int sched_show(struct seq_file *m, void *v)
1103 {
1104         struct inode *inode = m->private;
1105         struct task_struct *p;
1106
1107         WARN_ON(!inode);
1108
1109         p = get_proc_task(inode);
1110         if (!p)
1111                 return -ESRCH;
1112         proc_sched_show_task(p, m);
1113
1114         put_task_struct(p);
1115
1116         return 0;
1117 }
1118
1119 static ssize_t
1120 sched_write(struct file *file, const char __user *buf,
1121             size_t count, loff_t *offset)
1122 {
1123         struct inode *inode = file->f_path.dentry->d_inode;
1124         struct task_struct *p;
1125
1126         WARN_ON(!inode);
1127
1128         p = get_proc_task(inode);
1129         if (!p)
1130                 return -ESRCH;
1131         proc_sched_set_task(p);
1132
1133         put_task_struct(p);
1134
1135         return count;
1136 }
1137
1138 static int sched_open(struct inode *inode, struct file *filp)
1139 {
1140         int ret;
1141
1142         ret = single_open(filp, sched_show, NULL);
1143         if (!ret) {
1144                 struct seq_file *m = filp->private_data;
1145
1146                 m->private = inode;
1147         }
1148         return ret;
1149 }
1150
1151 static const struct file_operations proc_pid_sched_operations = {
1152         .open           = sched_open,
1153         .read           = seq_read,
1154         .write          = sched_write,
1155         .llseek         = seq_lseek,
1156         .release        = single_release,
1157 };
1158
1159 #endif
1160
1161 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1162 {
1163         struct inode *inode = dentry->d_inode;
1164         int error = -EACCES;
1165
1166         /* We don't need a base pointer in the /proc filesystem */
1167         path_put(&nd->path);
1168
1169         /* Are we allowed to snoop on the tasks file descriptors? */
1170         if (!proc_fd_access_allowed(inode))
1171                 goto out;
1172
1173         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1174         nd->last_type = LAST_BIND;
1175 out:
1176         return ERR_PTR(error);
1177 }
1178
1179 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1180 {
1181         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1182         char *pathname;
1183         int len;
1184
1185         if (!tmp)
1186                 return -ENOMEM;
1187
1188         pathname = d_path(path, tmp, PAGE_SIZE);
1189         len = PTR_ERR(pathname);
1190         if (IS_ERR(pathname))
1191                 goto out;
1192         len = tmp + PAGE_SIZE - 1 - pathname;
1193
1194         if (len > buflen)
1195                 len = buflen;
1196         if (copy_to_user(buffer, pathname, len))
1197                 len = -EFAULT;
1198  out:
1199         free_page((unsigned long)tmp);
1200         return len;
1201 }
1202
1203 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1204 {
1205         int error = -EACCES;
1206         struct inode *inode = dentry->d_inode;
1207         struct path path;
1208
1209         /* Are we allowed to snoop on the tasks file descriptors? */
1210         if (!proc_fd_access_allowed(inode))
1211                 goto out;
1212
1213         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1214         if (error)
1215                 goto out;
1216
1217         error = do_proc_readlink(&path, buffer, buflen);
1218         path_put(&path);
1219 out:
1220         return error;
1221 }
1222
1223 static const struct inode_operations proc_pid_link_inode_operations = {
1224         .readlink       = proc_pid_readlink,
1225         .follow_link    = proc_pid_follow_link,
1226         .setattr        = proc_setattr,
1227 };
1228
1229
1230 /* building an inode */
1231
1232 static int task_dumpable(struct task_struct *task)
1233 {
1234         int dumpable = 0;
1235         struct mm_struct *mm;
1236
1237         task_lock(task);
1238         mm = task->mm;
1239         if (mm)
1240                 dumpable = get_dumpable(mm);
1241         task_unlock(task);
1242         if(dumpable == 1)
1243                 return 1;
1244         return 0;
1245 }
1246
1247
1248 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1249 {
1250         struct inode * inode;
1251         struct proc_inode *ei;
1252
1253         /* We need a new inode */
1254
1255         inode = new_inode(sb);
1256         if (!inode)
1257                 goto out;
1258
1259         /* Common stuff */
1260         ei = PROC_I(inode);
1261         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1262         inode->i_op = &proc_def_inode_operations;
1263
1264         /*
1265          * grab the reference to task.
1266          */
1267         ei->pid = get_task_pid(task, PIDTYPE_PID);
1268         if (!ei->pid)
1269                 goto out_unlock;
1270
1271         inode->i_uid = 0;
1272         inode->i_gid = 0;
1273         if (task_dumpable(task)) {
1274                 inode->i_uid = task->euid;
1275                 inode->i_gid = task->egid;
1276         }
1277         security_task_to_inode(task, inode);
1278
1279 out:
1280         return inode;
1281
1282 out_unlock:
1283         iput(inode);
1284         return NULL;
1285 }
1286
1287 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1288 {
1289         struct inode *inode = dentry->d_inode;
1290         struct task_struct *task;
1291         generic_fillattr(inode, stat);
1292
1293         rcu_read_lock();
1294         stat->uid = 0;
1295         stat->gid = 0;
1296         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1297         if (task) {
1298                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1299                     task_dumpable(task)) {
1300                         stat->uid = task->euid;
1301                         stat->gid = task->egid;
1302                 }
1303         }
1304         rcu_read_unlock();
1305         return 0;
1306 }
1307
1308 /* dentry stuff */
1309
1310 /*
1311  *      Exceptional case: normally we are not allowed to unhash a busy
1312  * directory. In this case, however, we can do it - no aliasing problems
1313  * due to the way we treat inodes.
1314  *
1315  * Rewrite the inode's ownerships here because the owning task may have
1316  * performed a setuid(), etc.
1317  *
1318  * Before the /proc/pid/status file was created the only way to read
1319  * the effective uid of a /process was to stat /proc/pid.  Reading
1320  * /proc/pid/status is slow enough that procps and other packages
1321  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1322  * made this apply to all per process world readable and executable
1323  * directories.
1324  */
1325 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1326 {
1327         struct inode *inode = dentry->d_inode;
1328         struct task_struct *task = get_proc_task(inode);
1329         if (task) {
1330                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1331                     task_dumpable(task)) {
1332                         inode->i_uid = task->euid;
1333                         inode->i_gid = task->egid;
1334                 } else {
1335                         inode->i_uid = 0;
1336                         inode->i_gid = 0;
1337                 }
1338                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1339                 security_task_to_inode(task, inode);
1340                 put_task_struct(task);
1341                 return 1;
1342         }
1343         d_drop(dentry);
1344         return 0;
1345 }
1346
1347 static int pid_delete_dentry(struct dentry * dentry)
1348 {
1349         /* Is the task we represent dead?
1350          * If so, then don't put the dentry on the lru list,
1351          * kill it immediately.
1352          */
1353         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1354 }
1355
1356 static struct dentry_operations pid_dentry_operations =
1357 {
1358         .d_revalidate   = pid_revalidate,
1359         .d_delete       = pid_delete_dentry,
1360 };
1361
1362 /* Lookups */
1363
1364 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1365                                 struct task_struct *, const void *);
1366
1367 /*
1368  * Fill a directory entry.
1369  *
1370  * If possible create the dcache entry and derive our inode number and
1371  * file type from dcache entry.
1372  *
1373  * Since all of the proc inode numbers are dynamically generated, the inode
1374  * numbers do not exist until the inode is cache.  This means creating the
1375  * the dcache entry in readdir is necessary to keep the inode numbers
1376  * reported by readdir in sync with the inode numbers reported
1377  * by stat.
1378  */
1379 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1380         char *name, int len,
1381         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1382 {
1383         struct dentry *child, *dir = filp->f_path.dentry;
1384         struct inode *inode;
1385         struct qstr qname;
1386         ino_t ino = 0;
1387         unsigned type = DT_UNKNOWN;
1388
1389         qname.name = name;
1390         qname.len  = len;
1391         qname.hash = full_name_hash(name, len);
1392
1393         child = d_lookup(dir, &qname);
1394         if (!child) {
1395                 struct dentry *new;
1396                 new = d_alloc(dir, &qname);
1397                 if (new) {
1398                         child = instantiate(dir->d_inode, new, task, ptr);
1399                         if (child)
1400                                 dput(new);
1401                         else
1402                                 child = new;
1403                 }
1404         }
1405         if (!child || IS_ERR(child) || !child->d_inode)
1406                 goto end_instantiate;
1407         inode = child->d_inode;
1408         if (inode) {
1409                 ino = inode->i_ino;
1410                 type = inode->i_mode >> 12;
1411         }
1412         dput(child);
1413 end_instantiate:
1414         if (!ino)
1415                 ino = find_inode_number(dir, &qname);
1416         if (!ino)
1417                 ino = 1;
1418         return filldir(dirent, name, len, filp->f_pos, ino, type);
1419 }
1420
1421 static unsigned name_to_int(struct dentry *dentry)
1422 {
1423         const char *name = dentry->d_name.name;
1424         int len = dentry->d_name.len;
1425         unsigned n = 0;
1426
1427         if (len > 1 && *name == '0')
1428                 goto out;
1429         while (len-- > 0) {
1430                 unsigned c = *name++ - '0';
1431                 if (c > 9)
1432                         goto out;
1433                 if (n >= (~0U-9)/10)
1434                         goto out;
1435                 n *= 10;
1436                 n += c;
1437         }
1438         return n;
1439 out:
1440         return ~0U;
1441 }
1442
1443 #define PROC_FDINFO_MAX 64
1444
1445 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1446 {
1447         struct task_struct *task = get_proc_task(inode);
1448         struct files_struct *files = NULL;
1449         struct file *file;
1450         int fd = proc_fd(inode);
1451
1452         if (task) {
1453                 files = get_files_struct(task);
1454                 put_task_struct(task);
1455         }
1456         if (files) {
1457                 /*
1458                  * We are not taking a ref to the file structure, so we must
1459                  * hold ->file_lock.
1460                  */
1461                 spin_lock(&files->file_lock);
1462                 file = fcheck_files(files, fd);
1463                 if (file) {
1464                         if (path) {
1465                                 *path = file->f_path;
1466                                 path_get(&file->f_path);
1467                         }
1468                         if (info)
1469                                 snprintf(info, PROC_FDINFO_MAX,
1470                                          "pos:\t%lli\n"
1471                                          "flags:\t0%o\n",
1472                                          (long long) file->f_pos,
1473                                          file->f_flags);
1474                         spin_unlock(&files->file_lock);
1475                         put_files_struct(files);
1476                         return 0;
1477                 }
1478                 spin_unlock(&files->file_lock);
1479                 put_files_struct(files);
1480         }
1481         return -ENOENT;
1482 }
1483
1484 static int proc_fd_link(struct inode *inode, struct path *path)
1485 {
1486         return proc_fd_info(inode, path, NULL);
1487 }
1488
1489 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1490 {
1491         struct inode *inode = dentry->d_inode;
1492         struct task_struct *task = get_proc_task(inode);
1493         int fd = proc_fd(inode);
1494         struct files_struct *files;
1495
1496         if (task) {
1497                 files = get_files_struct(task);
1498                 if (files) {
1499                         rcu_read_lock();
1500                         if (fcheck_files(files, fd)) {
1501                                 rcu_read_unlock();
1502                                 put_files_struct(files);
1503                                 if (task_dumpable(task)) {
1504                                         inode->i_uid = task->euid;
1505                                         inode->i_gid = task->egid;
1506                                 } else {
1507                                         inode->i_uid = 0;
1508                                         inode->i_gid = 0;
1509                                 }
1510                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1511                                 security_task_to_inode(task, inode);
1512                                 put_task_struct(task);
1513                                 return 1;
1514                         }
1515                         rcu_read_unlock();
1516                         put_files_struct(files);
1517                 }
1518                 put_task_struct(task);
1519         }
1520         d_drop(dentry);
1521         return 0;
1522 }
1523
1524 static struct dentry_operations tid_fd_dentry_operations =
1525 {
1526         .d_revalidate   = tid_fd_revalidate,
1527         .d_delete       = pid_delete_dentry,
1528 };
1529
1530 static struct dentry *proc_fd_instantiate(struct inode *dir,
1531         struct dentry *dentry, struct task_struct *task, const void *ptr)
1532 {
1533         unsigned fd = *(const unsigned *)ptr;
1534         struct file *file;
1535         struct files_struct *files;
1536         struct inode *inode;
1537         struct proc_inode *ei;
1538         struct dentry *error = ERR_PTR(-ENOENT);
1539
1540         inode = proc_pid_make_inode(dir->i_sb, task);
1541         if (!inode)
1542                 goto out;
1543         ei = PROC_I(inode);
1544         ei->fd = fd;
1545         files = get_files_struct(task);
1546         if (!files)
1547                 goto out_iput;
1548         inode->i_mode = S_IFLNK;
1549
1550         /*
1551          * We are not taking a ref to the file structure, so we must
1552          * hold ->file_lock.
1553          */
1554         spin_lock(&files->file_lock);
1555         file = fcheck_files(files, fd);
1556         if (!file)
1557                 goto out_unlock;
1558         if (file->f_mode & 1)
1559                 inode->i_mode |= S_IRUSR | S_IXUSR;
1560         if (file->f_mode & 2)
1561                 inode->i_mode |= S_IWUSR | S_IXUSR;
1562         spin_unlock(&files->file_lock);
1563         put_files_struct(files);
1564
1565         inode->i_op = &proc_pid_link_inode_operations;
1566         inode->i_size = 64;
1567         ei->op.proc_get_link = proc_fd_link;
1568         dentry->d_op = &tid_fd_dentry_operations;
1569         d_add(dentry, inode);
1570         /* Close the race of the process dying before we return the dentry */
1571         if (tid_fd_revalidate(dentry, NULL))
1572                 error = NULL;
1573
1574  out:
1575         return error;
1576 out_unlock:
1577         spin_unlock(&files->file_lock);
1578         put_files_struct(files);
1579 out_iput:
1580         iput(inode);
1581         goto out;
1582 }
1583
1584 static struct dentry *proc_lookupfd_common(struct inode *dir,
1585                                            struct dentry *dentry,
1586                                            instantiate_t instantiate)
1587 {
1588         struct task_struct *task = get_proc_task(dir);
1589         unsigned fd = name_to_int(dentry);
1590         struct dentry *result = ERR_PTR(-ENOENT);
1591
1592         if (!task)
1593                 goto out_no_task;
1594         if (fd == ~0U)
1595                 goto out;
1596
1597         result = instantiate(dir, dentry, task, &fd);
1598 out:
1599         put_task_struct(task);
1600 out_no_task:
1601         return result;
1602 }
1603
1604 static int proc_readfd_common(struct file * filp, void * dirent,
1605                               filldir_t filldir, instantiate_t instantiate)
1606 {
1607         struct dentry *dentry = filp->f_path.dentry;
1608         struct inode *inode = dentry->d_inode;
1609         struct task_struct *p = get_proc_task(inode);
1610         unsigned int fd, ino;
1611         int retval;
1612         struct files_struct * files;
1613         struct fdtable *fdt;
1614
1615         retval = -ENOENT;
1616         if (!p)
1617                 goto out_no_task;
1618         retval = 0;
1619
1620         fd = filp->f_pos;
1621         switch (fd) {
1622                 case 0:
1623                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1624                                 goto out;
1625                         filp->f_pos++;
1626                 case 1:
1627                         ino = parent_ino(dentry);
1628                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1629                                 goto out;
1630                         filp->f_pos++;
1631                 default:
1632                         files = get_files_struct(p);
1633                         if (!files)
1634                                 goto out;
1635                         rcu_read_lock();
1636                         fdt = files_fdtable(files);
1637                         for (fd = filp->f_pos-2;
1638                              fd < fdt->max_fds;
1639                              fd++, filp->f_pos++) {
1640                                 char name[PROC_NUMBUF];
1641                                 int len;
1642
1643                                 if (!fcheck_files(files, fd))
1644                                         continue;
1645                                 rcu_read_unlock();
1646
1647                                 len = snprintf(name, sizeof(name), "%d", fd);
1648                                 if (proc_fill_cache(filp, dirent, filldir,
1649                                                     name, len, instantiate,
1650                                                     p, &fd) < 0) {
1651                                         rcu_read_lock();
1652                                         break;
1653                                 }
1654                                 rcu_read_lock();
1655                         }
1656                         rcu_read_unlock();
1657                         put_files_struct(files);
1658         }
1659 out:
1660         put_task_struct(p);
1661 out_no_task:
1662         return retval;
1663 }
1664
1665 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1666                                     struct nameidata *nd)
1667 {
1668         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1669 }
1670
1671 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1672 {
1673         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1674 }
1675
1676 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1677                                       size_t len, loff_t *ppos)
1678 {
1679         char tmp[PROC_FDINFO_MAX];
1680         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1681         if (!err)
1682                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1683         return err;
1684 }
1685
1686 static const struct file_operations proc_fdinfo_file_operations = {
1687         .open           = nonseekable_open,
1688         .read           = proc_fdinfo_read,
1689 };
1690
1691 static const struct file_operations proc_fd_operations = {
1692         .read           = generic_read_dir,
1693         .readdir        = proc_readfd,
1694 };
1695
1696 /*
1697  * /proc/pid/fd needs a special permission handler so that a process can still
1698  * access /proc/self/fd after it has executed a setuid().
1699  */
1700 static int proc_fd_permission(struct inode *inode, int mask,
1701                                 struct nameidata *nd)
1702 {
1703         int rv;
1704
1705         rv = generic_permission(inode, mask, NULL);
1706         if (rv == 0)
1707                 return 0;
1708         if (task_pid(current) == proc_pid(inode))
1709                 rv = 0;
1710         return rv;
1711 }
1712
1713 /*
1714  * proc directories can do almost nothing..
1715  */
1716 static const struct inode_operations proc_fd_inode_operations = {
1717         .lookup         = proc_lookupfd,
1718         .permission     = proc_fd_permission,
1719         .setattr        = proc_setattr,
1720 };
1721
1722 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1723         struct dentry *dentry, struct task_struct *task, const void *ptr)
1724 {
1725         unsigned fd = *(unsigned *)ptr;
1726         struct inode *inode;
1727         struct proc_inode *ei;
1728         struct dentry *error = ERR_PTR(-ENOENT);
1729
1730         inode = proc_pid_make_inode(dir->i_sb, task);
1731         if (!inode)
1732                 goto out;
1733         ei = PROC_I(inode);
1734         ei->fd = fd;
1735         inode->i_mode = S_IFREG | S_IRUSR;
1736         inode->i_fop = &proc_fdinfo_file_operations;
1737         dentry->d_op = &tid_fd_dentry_operations;
1738         d_add(dentry, inode);
1739         /* Close the race of the process dying before we return the dentry */
1740         if (tid_fd_revalidate(dentry, NULL))
1741                 error = NULL;
1742
1743  out:
1744         return error;
1745 }
1746
1747 static struct dentry *proc_lookupfdinfo(struct inode *dir,
1748                                         struct dentry *dentry,
1749                                         struct nameidata *nd)
1750 {
1751         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
1752 }
1753
1754 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
1755 {
1756         return proc_readfd_common(filp, dirent, filldir,
1757                                   proc_fdinfo_instantiate);
1758 }
1759
1760 static const struct file_operations proc_fdinfo_operations = {
1761         .read           = generic_read_dir,
1762         .readdir        = proc_readfdinfo,
1763 };
1764
1765 /*
1766  * proc directories can do almost nothing..
1767  */
1768 static const struct inode_operations proc_fdinfo_inode_operations = {
1769         .lookup         = proc_lookupfdinfo,
1770         .setattr        = proc_setattr,
1771 };
1772
1773
1774 static struct dentry *proc_pident_instantiate(struct inode *dir,
1775         struct dentry *dentry, struct task_struct *task, const void *ptr)
1776 {
1777         const struct pid_entry *p = ptr;
1778         struct inode *inode;
1779         struct proc_inode *ei;
1780         struct dentry *error = ERR_PTR(-EINVAL);
1781
1782         inode = proc_pid_make_inode(dir->i_sb, task);
1783         if (!inode)
1784                 goto out;
1785
1786         ei = PROC_I(inode);
1787         inode->i_mode = p->mode;
1788         if (S_ISDIR(inode->i_mode))
1789                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
1790         if (p->iop)
1791                 inode->i_op = p->iop;
1792         if (p->fop)
1793                 inode->i_fop = p->fop;
1794         ei->op = p->op;
1795         dentry->d_op = &pid_dentry_operations;
1796         d_add(dentry, inode);
1797         /* Close the race of the process dying before we return the dentry */
1798         if (pid_revalidate(dentry, NULL))
1799                 error = NULL;
1800 out:
1801         return error;
1802 }
1803
1804 static struct dentry *proc_pident_lookup(struct inode *dir,
1805                                          struct dentry *dentry,
1806                                          const struct pid_entry *ents,
1807                                          unsigned int nents)
1808 {
1809         struct inode *inode;
1810         struct dentry *error;
1811         struct task_struct *task = get_proc_task(dir);
1812         const struct pid_entry *p, *last;
1813
1814         error = ERR_PTR(-ENOENT);
1815         inode = NULL;
1816
1817         if (!task)
1818                 goto out_no_task;
1819
1820         /*
1821          * Yes, it does not scale. And it should not. Don't add
1822          * new entries into /proc/<tgid>/ without very good reasons.
1823          */
1824         last = &ents[nents - 1];
1825         for (p = ents; p <= last; p++) {
1826                 if (p->len != dentry->d_name.len)
1827                         continue;
1828                 if (!memcmp(dentry->d_name.name, p->name, p->len))
1829                         break;
1830         }
1831         if (p > last)
1832                 goto out;
1833
1834         error = proc_pident_instantiate(dir, dentry, task, p);
1835 out:
1836         put_task_struct(task);
1837 out_no_task:
1838         return error;
1839 }
1840
1841 static int proc_pident_fill_cache(struct file *filp, void *dirent,
1842         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
1843 {
1844         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
1845                                 proc_pident_instantiate, task, p);
1846 }
1847
1848 static int proc_pident_readdir(struct file *filp,
1849                 void *dirent, filldir_t filldir,
1850                 const struct pid_entry *ents, unsigned int nents)
1851 {
1852         int i;
1853         struct dentry *dentry = filp->f_path.dentry;
1854         struct inode *inode = dentry->d_inode;
1855         struct task_struct *task = get_proc_task(inode);
1856         const struct pid_entry *p, *last;
1857         ino_t ino;
1858         int ret;
1859
1860         ret = -ENOENT;
1861         if (!task)
1862                 goto out_no_task;
1863
1864         ret = 0;
1865         i = filp->f_pos;
1866         switch (i) {
1867         case 0:
1868                 ino = inode->i_ino;
1869                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
1870                         goto out;
1871                 i++;
1872                 filp->f_pos++;
1873                 /* fall through */
1874         case 1:
1875                 ino = parent_ino(dentry);
1876                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
1877                         goto out;
1878                 i++;
1879                 filp->f_pos++;
1880                 /* fall through */
1881         default:
1882                 i -= 2;
1883                 if (i >= nents) {
1884                         ret = 1;
1885                         goto out;
1886                 }
1887                 p = ents + i;
1888                 last = &ents[nents - 1];
1889                 while (p <= last) {
1890                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
1891                                 goto out;
1892                         filp->f_pos++;
1893                         p++;
1894                 }
1895         }
1896
1897         ret = 1;
1898 out:
1899         put_task_struct(task);
1900 out_no_task:
1901         return ret;
1902 }
1903
1904 #ifdef CONFIG_SECURITY
1905 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
1906                                   size_t count, loff_t *ppos)
1907 {
1908         struct inode * inode = file->f_path.dentry->d_inode;
1909         char *p = NULL;
1910         ssize_t length;
1911         struct task_struct *task = get_proc_task(inode);
1912
1913         if (!task)
1914                 return -ESRCH;
1915
1916         length = security_getprocattr(task,
1917                                       (char*)file->f_path.dentry->d_name.name,
1918                                       &p);
1919         put_task_struct(task);
1920         if (length > 0)
1921                 length = simple_read_from_buffer(buf, count, ppos, p, length);
1922         kfree(p);
1923         return length;
1924 }
1925
1926 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
1927                                    size_t count, loff_t *ppos)
1928 {
1929         struct inode * inode = file->f_path.dentry->d_inode;
1930         char *page;
1931         ssize_t length;
1932         struct task_struct *task = get_proc_task(inode);
1933
1934         length = -ESRCH;
1935         if (!task)
1936                 goto out_no_task;
1937         if (count > PAGE_SIZE)
1938                 count = PAGE_SIZE;
1939
1940         /* No partial writes. */
1941         length = -EINVAL;
1942         if (*ppos != 0)
1943                 goto out;
1944
1945         length = -ENOMEM;
1946         page = (char*)__get_free_page(GFP_TEMPORARY);
1947         if (!page)
1948                 goto out;
1949
1950         length = -EFAULT;
1951         if (copy_from_user(page, buf, count))
1952                 goto out_free;
1953
1954         length = security_setprocattr(task,
1955                                       (char*)file->f_path.dentry->d_name.name,
1956                                       (void*)page, count);
1957 out_free:
1958         free_page((unsigned long) page);
1959 out:
1960         put_task_struct(task);
1961 out_no_task:
1962         return length;
1963 }
1964
1965 static const struct file_operations proc_pid_attr_operations = {
1966         .read           = proc_pid_attr_read,
1967         .write          = proc_pid_attr_write,
1968 };
1969
1970 static const struct pid_entry attr_dir_stuff[] = {
1971         REG("current",    S_IRUGO|S_IWUGO, pid_attr),
1972         REG("prev",       S_IRUGO,         pid_attr),
1973         REG("exec",       S_IRUGO|S_IWUGO, pid_attr),
1974         REG("fscreate",   S_IRUGO|S_IWUGO, pid_attr),
1975         REG("keycreate",  S_IRUGO|S_IWUGO, pid_attr),
1976         REG("sockcreate", S_IRUGO|S_IWUGO, pid_attr),
1977 };
1978
1979 static int proc_attr_dir_readdir(struct file * filp,
1980                              void * dirent, filldir_t filldir)
1981 {
1982         return proc_pident_readdir(filp,dirent,filldir,
1983                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
1984 }
1985
1986 static const struct file_operations proc_attr_dir_operations = {
1987         .read           = generic_read_dir,
1988         .readdir        = proc_attr_dir_readdir,
1989 };
1990
1991 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
1992                                 struct dentry *dentry, struct nameidata *nd)
1993 {
1994         return proc_pident_lookup(dir, dentry,
1995                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
1996 }
1997
1998 static const struct inode_operations proc_attr_dir_inode_operations = {
1999         .lookup         = proc_attr_dir_lookup,
2000         .getattr        = pid_getattr,
2001         .setattr        = proc_setattr,
2002 };
2003
2004 #endif
2005
2006 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2007 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2008                                          size_t count, loff_t *ppos)
2009 {
2010         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2011         struct mm_struct *mm;
2012         char buffer[PROC_NUMBUF];
2013         size_t len;
2014         int ret;
2015
2016         if (!task)
2017                 return -ESRCH;
2018
2019         ret = 0;
2020         mm = get_task_mm(task);
2021         if (mm) {
2022                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2023                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2024                                 MMF_DUMP_FILTER_SHIFT));
2025                 mmput(mm);
2026                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2027         }
2028
2029         put_task_struct(task);
2030
2031         return ret;
2032 }
2033
2034 static ssize_t proc_coredump_filter_write(struct file *file,
2035                                           const char __user *buf,
2036                                           size_t count,
2037                                           loff_t *ppos)
2038 {
2039         struct task_struct *task;
2040         struct mm_struct *mm;
2041         char buffer[PROC_NUMBUF], *end;
2042         unsigned int val;
2043         int ret;
2044         int i;
2045         unsigned long mask;
2046
2047         ret = -EFAULT;
2048         memset(buffer, 0, sizeof(buffer));
2049         if (count > sizeof(buffer) - 1)
2050                 count = sizeof(buffer) - 1;
2051         if (copy_from_user(buffer, buf, count))
2052                 goto out_no_task;
2053
2054         ret = -EINVAL;
2055         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2056         if (*end == '\n')
2057                 end++;
2058         if (end - buffer == 0)
2059                 goto out_no_task;
2060
2061         ret = -ESRCH;
2062         task = get_proc_task(file->f_dentry->d_inode);
2063         if (!task)
2064                 goto out_no_task;
2065
2066         ret = end - buffer;
2067         mm = get_task_mm(task);
2068         if (!mm)
2069                 goto out_no_mm;
2070
2071         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2072                 if (val & mask)
2073                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2074                 else
2075                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2076         }
2077
2078         mmput(mm);
2079  out_no_mm:
2080         put_task_struct(task);
2081  out_no_task:
2082         return ret;
2083 }
2084
2085 static const struct file_operations proc_coredump_filter_operations = {
2086         .read           = proc_coredump_filter_read,
2087         .write          = proc_coredump_filter_write,
2088 };
2089 #endif
2090
2091 /*
2092  * /proc/self:
2093  */
2094 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2095                               int buflen)
2096 {
2097         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2098         pid_t tgid = task_tgid_nr_ns(current, ns);
2099         char tmp[PROC_NUMBUF];
2100         if (!tgid)
2101                 return -ENOENT;
2102         sprintf(tmp, "%d", tgid);
2103         return vfs_readlink(dentry,buffer,buflen,tmp);
2104 }
2105
2106 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2107 {
2108         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2109         pid_t tgid = task_tgid_nr_ns(current, ns);
2110         char tmp[PROC_NUMBUF];
2111         if (!tgid)
2112                 return ERR_PTR(-ENOENT);
2113         sprintf(tmp, "%d", task_tgid_nr_ns(current, ns));
2114         return ERR_PTR(vfs_follow_link(nd,tmp));
2115 }
2116
2117 static const struct inode_operations proc_self_inode_operations = {
2118         .readlink       = proc_self_readlink,
2119         .follow_link    = proc_self_follow_link,
2120 };
2121
2122 /*
2123  * proc base
2124  *
2125  * These are the directory entries in the root directory of /proc
2126  * that properly belong to the /proc filesystem, as they describe
2127  * describe something that is process related.
2128  */
2129 static const struct pid_entry proc_base_stuff[] = {
2130         NOD("self", S_IFLNK|S_IRWXUGO,
2131                 &proc_self_inode_operations, NULL, {}),
2132 };
2133
2134 /*
2135  *      Exceptional case: normally we are not allowed to unhash a busy
2136  * directory. In this case, however, we can do it - no aliasing problems
2137  * due to the way we treat inodes.
2138  */
2139 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2140 {
2141         struct inode *inode = dentry->d_inode;
2142         struct task_struct *task = get_proc_task(inode);
2143         if (task) {
2144                 put_task_struct(task);
2145                 return 1;
2146         }
2147         d_drop(dentry);
2148         return 0;
2149 }
2150
2151 static struct dentry_operations proc_base_dentry_operations =
2152 {
2153         .d_revalidate   = proc_base_revalidate,
2154         .d_delete       = pid_delete_dentry,
2155 };
2156
2157 static struct dentry *proc_base_instantiate(struct inode *dir,
2158         struct dentry *dentry, struct task_struct *task, const void *ptr)
2159 {
2160         const struct pid_entry *p = ptr;
2161         struct inode *inode;
2162         struct proc_inode *ei;
2163         struct dentry *error = ERR_PTR(-EINVAL);
2164
2165         /* Allocate the inode */
2166         error = ERR_PTR(-ENOMEM);
2167         inode = new_inode(dir->i_sb);
2168         if (!inode)
2169                 goto out;
2170
2171         /* Initialize the inode */
2172         ei = PROC_I(inode);
2173         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2174
2175         /*
2176          * grab the reference to the task.
2177          */
2178         ei->pid = get_task_pid(task, PIDTYPE_PID);
2179         if (!ei->pid)
2180                 goto out_iput;
2181
2182         inode->i_uid = 0;
2183         inode->i_gid = 0;
2184         inode->i_mode = p->mode;
2185         if (S_ISDIR(inode->i_mode))
2186                 inode->i_nlink = 2;
2187         if (S_ISLNK(inode->i_mode))
2188                 inode->i_size = 64;
2189         if (p->iop)
2190                 inode->i_op = p->iop;
2191         if (p->fop)
2192                 inode->i_fop = p->fop;
2193         ei->op = p->op;
2194         dentry->d_op = &proc_base_dentry_operations;
2195         d_add(dentry, inode);
2196         error = NULL;
2197 out:
2198         return error;
2199 out_iput:
2200         iput(inode);
2201         goto out;
2202 }
2203
2204 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2205 {
2206         struct dentry *error;
2207         struct task_struct *task = get_proc_task(dir);
2208         const struct pid_entry *p, *last;
2209
2210         error = ERR_PTR(-ENOENT);
2211
2212         if (!task)
2213                 goto out_no_task;
2214
2215         /* Lookup the directory entry */
2216         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2217         for (p = proc_base_stuff; p <= last; p++) {
2218                 if (p->len != dentry->d_name.len)
2219                         continue;
2220                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2221                         break;
2222         }
2223         if (p > last)
2224                 goto out;
2225
2226         error = proc_base_instantiate(dir, dentry, task, p);
2227
2228 out:
2229         put_task_struct(task);
2230 out_no_task:
2231         return error;
2232 }
2233
2234 static int proc_base_fill_cache(struct file *filp, void *dirent,
2235         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2236 {
2237         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2238                                 proc_base_instantiate, task, p);
2239 }
2240
2241 #ifdef CONFIG_TASK_IO_ACCOUNTING
2242 static int proc_pid_io_accounting(struct task_struct *task, char *buffer)
2243 {
2244         return sprintf(buffer,
2245 #ifdef CONFIG_TASK_XACCT
2246                         "rchar: %llu\n"
2247                         "wchar: %llu\n"
2248                         "syscr: %llu\n"
2249                         "syscw: %llu\n"
2250 #endif
2251                         "read_bytes: %llu\n"
2252                         "write_bytes: %llu\n"
2253                         "cancelled_write_bytes: %llu\n",
2254 #ifdef CONFIG_TASK_XACCT
2255                         (unsigned long long)task->rchar,
2256                         (unsigned long long)task->wchar,
2257                         (unsigned long long)task->syscr,
2258                         (unsigned long long)task->syscw,
2259 #endif
2260                         (unsigned long long)task->ioac.read_bytes,
2261                         (unsigned long long)task->ioac.write_bytes,
2262                         (unsigned long long)task->ioac.cancelled_write_bytes);
2263 }
2264 #endif
2265
2266 /*
2267  * Thread groups
2268  */
2269 static const struct file_operations proc_task_operations;
2270 static const struct inode_operations proc_task_inode_operations;
2271
2272 static const struct pid_entry tgid_base_stuff[] = {
2273         DIR("task",       S_IRUGO|S_IXUGO, task),
2274         DIR("fd",         S_IRUSR|S_IXUSR, fd),
2275         DIR("fdinfo",     S_IRUSR|S_IXUSR, fdinfo),
2276         REG("environ",    S_IRUSR, environ),
2277         INF("auxv",       S_IRUSR, pid_auxv),
2278         ONE("status",     S_IRUGO, pid_status),
2279         INF("limits",     S_IRUSR, pid_limits),
2280 #ifdef CONFIG_SCHED_DEBUG
2281         REG("sched",      S_IRUGO|S_IWUSR, pid_sched),
2282 #endif
2283         INF("cmdline",    S_IRUGO, pid_cmdline),
2284         ONE("stat",       S_IRUGO, tgid_stat),
2285         ONE("statm",      S_IRUGO, pid_statm),
2286         REG("maps",       S_IRUGO, maps),
2287 #ifdef CONFIG_NUMA
2288         REG("numa_maps",  S_IRUGO, numa_maps),
2289 #endif
2290         REG("mem",        S_IRUSR|S_IWUSR, mem),
2291         LNK("cwd",        cwd),
2292         LNK("root",       root),
2293         LNK("exe",        exe),
2294         REG("mounts",     S_IRUGO, mounts),
2295         REG("mountstats", S_IRUSR, mountstats),
2296 #ifdef CONFIG_PROC_PAGE_MONITOR
2297         REG("clear_refs", S_IWUSR, clear_refs),
2298         REG("smaps",      S_IRUGO, smaps),
2299         REG("pagemap",    S_IRUSR, pagemap),
2300 #endif
2301 #ifdef CONFIG_SECURITY
2302         DIR("attr",       S_IRUGO|S_IXUGO, attr_dir),
2303 #endif
2304 #ifdef CONFIG_KALLSYMS
2305         INF("wchan",      S_IRUGO, pid_wchan),
2306 #endif
2307 #ifdef CONFIG_SCHEDSTATS
2308         INF("schedstat",  S_IRUGO, pid_schedstat),
2309 #endif
2310 #ifdef CONFIG_LATENCYTOP
2311         REG("latency",  S_IRUGO, lstats),
2312 #endif
2313 #ifdef CONFIG_PROC_PID_CPUSET
2314         REG("cpuset",     S_IRUGO, cpuset),
2315 #endif
2316 #ifdef CONFIG_CGROUPS
2317         REG("cgroup",  S_IRUGO, cgroup),
2318 #endif
2319         INF("oom_score",  S_IRUGO, oom_score),
2320         REG("oom_adj",    S_IRUGO|S_IWUSR, oom_adjust),
2321 #ifdef CONFIG_AUDITSYSCALL
2322         REG("loginuid",   S_IWUSR|S_IRUGO, loginuid),
2323 #endif
2324 #ifdef CONFIG_FAULT_INJECTION
2325         REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2326 #endif
2327 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2328         REG("coredump_filter", S_IRUGO|S_IWUSR, coredump_filter),
2329 #endif
2330 #ifdef CONFIG_TASK_IO_ACCOUNTING
2331         INF("io",       S_IRUGO, pid_io_accounting),
2332 #endif
2333 };
2334
2335 static int proc_tgid_base_readdir(struct file * filp,
2336                              void * dirent, filldir_t filldir)
2337 {
2338         return proc_pident_readdir(filp,dirent,filldir,
2339                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2340 }
2341
2342 static const struct file_operations proc_tgid_base_operations = {
2343         .read           = generic_read_dir,
2344         .readdir        = proc_tgid_base_readdir,
2345 };
2346
2347 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2348         return proc_pident_lookup(dir, dentry,
2349                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2350 }
2351
2352 static const struct inode_operations proc_tgid_base_inode_operations = {
2353         .lookup         = proc_tgid_base_lookup,
2354         .getattr        = pid_getattr,
2355         .setattr        = proc_setattr,
2356 };
2357
2358 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2359 {
2360         struct dentry *dentry, *leader, *dir;
2361         char buf[PROC_NUMBUF];
2362         struct qstr name;
2363
2364         name.name = buf;
2365         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2366         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2367         if (dentry) {
2368                 if (!(current->flags & PF_EXITING))
2369                         shrink_dcache_parent(dentry);
2370                 d_drop(dentry);
2371                 dput(dentry);
2372         }
2373
2374         if (tgid == 0)
2375                 goto out;
2376
2377         name.name = buf;
2378         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2379         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2380         if (!leader)
2381                 goto out;
2382
2383         name.name = "task";
2384         name.len = strlen(name.name);
2385         dir = d_hash_and_lookup(leader, &name);
2386         if (!dir)
2387                 goto out_put_leader;
2388
2389         name.name = buf;
2390         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2391         dentry = d_hash_and_lookup(dir, &name);
2392         if (dentry) {
2393                 shrink_dcache_parent(dentry);
2394                 d_drop(dentry);
2395                 dput(dentry);
2396         }
2397
2398         dput(dir);
2399 out_put_leader:
2400         dput(leader);
2401 out:
2402         return;
2403 }
2404
2405 /**
2406  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2407  * @task: task that should be flushed.
2408  *
2409  * When flushing dentries from proc, one needs to flush them from global
2410  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2411  * in. This call is supposed to do all of this job.
2412  *
2413  * Looks in the dcache for
2414  * /proc/@pid
2415  * /proc/@tgid/task/@pid
2416  * if either directory is present flushes it and all of it'ts children
2417  * from the dcache.
2418  *
2419  * It is safe and reasonable to cache /proc entries for a task until
2420  * that task exits.  After that they just clog up the dcache with
2421  * useless entries, possibly causing useful dcache entries to be
2422  * flushed instead.  This routine is proved to flush those useless
2423  * dcache entries at process exit time.
2424  *
2425  * NOTE: This routine is just an optimization so it does not guarantee
2426  *       that no dcache entries will exist at process exit time it
2427  *       just makes it very unlikely that any will persist.
2428  */
2429
2430 void proc_flush_task(struct task_struct *task)
2431 {
2432         int i;
2433         struct pid *pid, *tgid = NULL;
2434         struct upid *upid;
2435
2436         pid = task_pid(task);
2437         if (thread_group_leader(task))
2438                 tgid = task_tgid(task);
2439
2440         for (i = 0; i <= pid->level; i++) {
2441                 upid = &pid->numbers[i];
2442                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2443                         tgid ? tgid->numbers[i].nr : 0);
2444         }
2445
2446         upid = &pid->numbers[pid->level];
2447         if (upid->nr == 1)
2448                 pid_ns_release_proc(upid->ns);
2449 }
2450
2451 static struct dentry *proc_pid_instantiate(struct inode *dir,
2452                                            struct dentry * dentry,
2453                                            struct task_struct *task, const void *ptr)
2454 {
2455         struct dentry *error = ERR_PTR(-ENOENT);
2456         struct inode *inode;
2457
2458         inode = proc_pid_make_inode(dir->i_sb, task);
2459         if (!inode)
2460                 goto out;
2461
2462         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2463         inode->i_op = &proc_tgid_base_inode_operations;
2464         inode->i_fop = &proc_tgid_base_operations;
2465         inode->i_flags|=S_IMMUTABLE;
2466         inode->i_nlink = 5;
2467 #ifdef CONFIG_SECURITY
2468         inode->i_nlink += 1;
2469 #endif
2470
2471         dentry->d_op = &pid_dentry_operations;
2472
2473         d_add(dentry, inode);
2474         /* Close the race of the process dying before we return the dentry */
2475         if (pid_revalidate(dentry, NULL))
2476                 error = NULL;
2477 out:
2478         return error;
2479 }
2480
2481 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2482 {
2483         struct dentry *result = ERR_PTR(-ENOENT);
2484         struct task_struct *task;
2485         unsigned tgid;
2486         struct pid_namespace *ns;
2487
2488         result = proc_base_lookup(dir, dentry);
2489         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2490                 goto out;
2491
2492         tgid = name_to_int(dentry);
2493         if (tgid == ~0U)
2494                 goto out;
2495
2496         ns = dentry->d_sb->s_fs_info;
2497         rcu_read_lock();
2498         task = find_task_by_pid_ns(tgid, ns);
2499         if (task)
2500                 get_task_struct(task);
2501         rcu_read_unlock();
2502         if (!task)
2503                 goto out;
2504
2505         result = proc_pid_instantiate(dir, dentry, task, NULL);
2506         put_task_struct(task);
2507 out:
2508         return result;
2509 }
2510
2511 /*
2512  * Find the first task with tgid >= tgid
2513  *
2514  */
2515 struct tgid_iter {
2516         unsigned int tgid;
2517         struct task_struct *task;
2518 };
2519 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2520 {
2521         struct pid *pid;
2522
2523         if (iter.task)
2524                 put_task_struct(iter.task);
2525         rcu_read_lock();
2526 retry:
2527         iter.task = NULL;
2528         pid = find_ge_pid(iter.tgid, ns);
2529         if (pid) {
2530                 iter.tgid = pid_nr_ns(pid, ns);
2531                 iter.task = pid_task(pid, PIDTYPE_PID);
2532                 /* What we to know is if the pid we have find is the
2533                  * pid of a thread_group_leader.  Testing for task
2534                  * being a thread_group_leader is the obvious thing
2535                  * todo but there is a window when it fails, due to
2536                  * the pid transfer logic in de_thread.
2537                  *
2538                  * So we perform the straight forward test of seeing
2539                  * if the pid we have found is the pid of a thread
2540                  * group leader, and don't worry if the task we have
2541                  * found doesn't happen to be a thread group leader.
2542                  * As we don't care in the case of readdir.
2543                  */
2544                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2545                         iter.tgid += 1;
2546                         goto retry;
2547                 }
2548                 get_task_struct(iter.task);
2549         }
2550         rcu_read_unlock();
2551         return iter;
2552 }
2553
2554 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2555
2556 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2557         struct tgid_iter iter)
2558 {
2559         char name[PROC_NUMBUF];
2560         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2561         return proc_fill_cache(filp, dirent, filldir, name, len,
2562                                 proc_pid_instantiate, iter.task, NULL);
2563 }
2564
2565 /* for the /proc/ directory itself, after non-process stuff has been done */
2566 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2567 {
2568         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2569         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2570         struct tgid_iter iter;
2571         struct pid_namespace *ns;
2572
2573         if (!reaper)
2574                 goto out_no_task;
2575
2576         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2577                 const struct pid_entry *p = &proc_base_stuff[nr];
2578                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2579                         goto out;
2580         }
2581
2582         ns = filp->f_dentry->d_sb->s_fs_info;
2583         iter.task = NULL;
2584         iter.tgid = filp->f_pos - TGID_OFFSET;
2585         for (iter = next_tgid(ns, iter);
2586              iter.task;
2587              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2588                 filp->f_pos = iter.tgid + TGID_OFFSET;
2589                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2590                         put_task_struct(iter.task);
2591                         goto out;
2592                 }
2593         }
2594         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2595 out:
2596         put_task_struct(reaper);
2597 out_no_task:
2598         return 0;
2599 }
2600
2601 /*
2602  * Tasks
2603  */
2604 static const struct pid_entry tid_base_stuff[] = {
2605         DIR("fd",        S_IRUSR|S_IXUSR, fd),
2606         DIR("fdinfo",    S_IRUSR|S_IXUSR, fdinfo),
2607         REG("environ",   S_IRUSR, environ),
2608         INF("auxv",      S_IRUSR, pid_auxv),
2609         ONE("status",    S_IRUGO, pid_status),
2610         INF("limits",    S_IRUSR, pid_limits),
2611 #ifdef CONFIG_SCHED_DEBUG
2612         REG("sched",     S_IRUGO|S_IWUSR, pid_sched),
2613 #endif
2614         INF("cmdline",   S_IRUGO, pid_cmdline),
2615         ONE("stat",      S_IRUGO, tid_stat),
2616         ONE("statm",     S_IRUGO, pid_statm),
2617         REG("maps",      S_IRUGO, maps),
2618 #ifdef CONFIG_NUMA
2619         REG("numa_maps", S_IRUGO, numa_maps),
2620 #endif
2621         REG("mem",       S_IRUSR|S_IWUSR, mem),
2622         LNK("cwd",       cwd),
2623         LNK("root",      root),
2624         LNK("exe",       exe),
2625         REG("mounts",    S_IRUGO, mounts),
2626 #ifdef CONFIG_PROC_PAGE_MONITOR
2627         REG("clear_refs", S_IWUSR, clear_refs),
2628         REG("smaps",     S_IRUGO, smaps),
2629         REG("pagemap",    S_IRUSR, pagemap),
2630 #endif
2631 #ifdef CONFIG_SECURITY
2632         DIR("attr",      S_IRUGO|S_IXUGO, attr_dir),
2633 #endif
2634 #ifdef CONFIG_KALLSYMS
2635         INF("wchan",     S_IRUGO, pid_wchan),
2636 #endif
2637 #ifdef CONFIG_SCHEDSTATS
2638         INF("schedstat", S_IRUGO, pid_schedstat),
2639 #endif
2640 #ifdef CONFIG_LATENCYTOP
2641         REG("latency",  S_IRUGO, lstats),
2642 #endif
2643 #ifdef CONFIG_PROC_PID_CPUSET
2644         REG("cpuset",    S_IRUGO, cpuset),
2645 #endif
2646 #ifdef CONFIG_CGROUPS
2647         REG("cgroup",  S_IRUGO, cgroup),
2648 #endif
2649         INF("oom_score", S_IRUGO, oom_score),
2650         REG("oom_adj",   S_IRUGO|S_IWUSR, oom_adjust),
2651 #ifdef CONFIG_AUDITSYSCALL
2652         REG("loginuid",  S_IWUSR|S_IRUGO, loginuid),
2653 #endif
2654 #ifdef CONFIG_FAULT_INJECTION
2655         REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2656 #endif
2657 };
2658
2659 static int proc_tid_base_readdir(struct file * filp,
2660                              void * dirent, filldir_t filldir)
2661 {
2662         return proc_pident_readdir(filp,dirent,filldir,
2663                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2664 }
2665
2666 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2667         return proc_pident_lookup(dir, dentry,
2668                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2669 }
2670
2671 static const struct file_operations proc_tid_base_operations = {
2672         .read           = generic_read_dir,
2673         .readdir        = proc_tid_base_readdir,
2674 };
2675
2676 static const struct inode_operations proc_tid_base_inode_operations = {
2677         .lookup         = proc_tid_base_lookup,
2678         .getattr        = pid_getattr,
2679         .setattr        = proc_setattr,
2680 };
2681
2682 static struct dentry *proc_task_instantiate(struct inode *dir,
2683         struct dentry *dentry, struct task_struct *task, const void *ptr)
2684 {
2685         struct dentry *error = ERR_PTR(-ENOENT);
2686         struct inode *inode;
2687         inode = proc_pid_make_inode(dir->i_sb, task);
2688
2689         if (!inode)
2690                 goto out;
2691         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2692         inode->i_op = &proc_tid_base_inode_operations;
2693         inode->i_fop = &proc_tid_base_operations;
2694         inode->i_flags|=S_IMMUTABLE;
2695         inode->i_nlink = 4;
2696 #ifdef CONFIG_SECURITY
2697         inode->i_nlink += 1;
2698 #endif
2699
2700         dentry->d_op = &pid_dentry_operations;
2701
2702         d_add(dentry, inode);
2703         /* Close the race of the process dying before we return the dentry */
2704         if (pid_revalidate(dentry, NULL))
2705                 error = NULL;
2706 out:
2707         return error;
2708 }
2709
2710 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2711 {
2712         struct dentry *result = ERR_PTR(-ENOENT);
2713         struct task_struct *task;
2714         struct task_struct *leader = get_proc_task(dir);
2715         unsigned tid;
2716         struct pid_namespace *ns;
2717
2718         if (!leader)
2719                 goto out_no_task;
2720
2721         tid = name_to_int(dentry);
2722         if (tid == ~0U)
2723                 goto out;
2724
2725         ns = dentry->d_sb->s_fs_info;
2726         rcu_read_lock();
2727         task = find_task_by_pid_ns(tid, ns);
2728         if (task)
2729                 get_task_struct(task);
2730         rcu_read_unlock();
2731         if (!task)
2732                 goto out;
2733         if (!same_thread_group(leader, task))
2734                 goto out_drop_task;
2735
2736         result = proc_task_instantiate(dir, dentry, task, NULL);
2737 out_drop_task:
2738         put_task_struct(task);
2739 out:
2740         put_task_struct(leader);
2741 out_no_task:
2742         return result;
2743 }
2744
2745 /*
2746  * Find the first tid of a thread group to return to user space.
2747  *
2748  * Usually this is just the thread group leader, but if the users
2749  * buffer was too small or there was a seek into the middle of the
2750  * directory we have more work todo.
2751  *
2752  * In the case of a short read we start with find_task_by_pid.
2753  *
2754  * In the case of a seek we start with the leader and walk nr
2755  * threads past it.
2756  */
2757 static struct task_struct *first_tid(struct task_struct *leader,
2758                 int tid, int nr, struct pid_namespace *ns)
2759 {
2760         struct task_struct *pos;
2761
2762         rcu_read_lock();
2763         /* Attempt to start with the pid of a thread */
2764         if (tid && (nr > 0)) {
2765                 pos = find_task_by_pid_ns(tid, ns);
2766                 if (pos && (pos->group_leader == leader))
2767                         goto found;
2768         }
2769
2770         /* If nr exceeds the number of threads there is nothing todo */
2771         pos = NULL;
2772         if (nr && nr >= get_nr_threads(leader))
2773                 goto out;
2774
2775         /* If we haven't found our starting place yet start
2776          * with the leader and walk nr threads forward.
2777          */
2778         for (pos = leader; nr > 0; --nr) {
2779                 pos = next_thread(pos);
2780                 if (pos == leader) {
2781                         pos = NULL;
2782                         goto out;
2783                 }
2784         }
2785 found:
2786         get_task_struct(pos);
2787 out:
2788         rcu_read_unlock();
2789         return pos;
2790 }
2791
2792 /*
2793  * Find the next thread in the thread list.
2794  * Return NULL if there is an error or no next thread.
2795  *
2796  * The reference to the input task_struct is released.
2797  */
2798 static struct task_struct *next_tid(struct task_struct *start)
2799 {
2800         struct task_struct *pos = NULL;
2801         rcu_read_lock();
2802         if (pid_alive(start)) {
2803                 pos = next_thread(start);
2804                 if (thread_group_leader(pos))
2805                         pos = NULL;
2806                 else
2807                         get_task_struct(pos);
2808         }
2809         rcu_read_unlock();
2810         put_task_struct(start);
2811         return pos;
2812 }
2813
2814 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2815         struct task_struct *task, int tid)
2816 {
2817         char name[PROC_NUMBUF];
2818         int len = snprintf(name, sizeof(name), "%d", tid);
2819         return proc_fill_cache(filp, dirent, filldir, name, len,
2820                                 proc_task_instantiate, task, NULL);
2821 }
2822
2823 /* for the /proc/TGID/task/ directories */
2824 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
2825 {
2826         struct dentry *dentry = filp->f_path.dentry;
2827         struct inode *inode = dentry->d_inode;
2828         struct task_struct *leader = NULL;
2829         struct task_struct *task;
2830         int retval = -ENOENT;
2831         ino_t ino;
2832         int tid;
2833         unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
2834         struct pid_namespace *ns;
2835
2836         task = get_proc_task(inode);
2837         if (!task)
2838                 goto out_no_task;
2839         rcu_read_lock();
2840         if (pid_alive(task)) {
2841                 leader = task->group_leader;
2842                 get_task_struct(leader);
2843         }
2844         rcu_read_unlock();
2845         put_task_struct(task);
2846         if (!leader)
2847                 goto out_no_task;
2848         retval = 0;
2849
2850         switch (pos) {
2851         case 0:
2852                 ino = inode->i_ino;
2853                 if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
2854                         goto out;
2855                 pos++;
2856                 /* fall through */
2857         case 1:
2858                 ino = parent_ino(dentry);
2859                 if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
2860                         goto out;
2861                 pos++;
2862                 /* fall through */
2863         }
2864
2865         /* f_version caches the tgid value that the last readdir call couldn't
2866          * return. lseek aka telldir automagically resets f_version to 0.
2867          */
2868         ns = filp->f_dentry->d_sb->s_fs_info;
2869         tid = (int)filp->f_version;
2870         filp->f_version = 0;
2871         for (task = first_tid(leader, tid, pos - 2, ns);
2872              task;
2873              task = next_tid(task), pos++) {
2874                 tid = task_pid_nr_ns(task, ns);
2875                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
2876                         /* returning this tgid failed, save it as the first
2877                          * pid for the next readir call */
2878                         filp->f_version = (u64)tid;
2879                         put_task_struct(task);
2880                         break;
2881                 }
2882         }
2883 out:
2884         filp->f_pos = pos;
2885         put_task_struct(leader);
2886 out_no_task:
2887         return retval;
2888 }
2889
2890 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
2891 {
2892         struct inode *inode = dentry->d_inode;
2893         struct task_struct *p = get_proc_task(inode);
2894         generic_fillattr(inode, stat);
2895
2896         if (p) {
2897                 rcu_read_lock();
2898                 stat->nlink += get_nr_threads(p);
2899                 rcu_read_unlock();
2900                 put_task_struct(p);
2901         }
2902
2903         return 0;
2904 }
2905
2906 static const struct inode_operations proc_task_inode_operations = {
2907         .lookup         = proc_task_lookup,
2908         .getattr        = proc_task_getattr,
2909         .setattr        = proc_setattr,
2910 };
2911
2912 static const struct file_operations proc_task_operations = {
2913         .read           = generic_read_dir,
2914         .readdir        = proc_task_readdir,
2915 };