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