SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/proc/base.c

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