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