fs/proc/base.c

   1 /*
   2  *  linux/fs/proc/base.c
   3  *
   4  *  Copyright (C) 1991, 1992 Linus Torvalds
   5  *
   6  *  proc base directory handling functions
   7  *
   8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
   9  *  Instead of using magical inumbers to determine the kind of object
  10  *  we allocate and fill in-core inodes upon lookup. They don't even
  11  *  go into icache. We cache the reference to task_struct upon lookup too.
  12  *  Eventually it should become a filesystem in its own. We don't use the
  13  *  rest of procfs anymore.
  14  *
  15  *
  16  *  Changelog:
  17  *  17-Jan-2005
  18  *  Allan Bezerra
  19  *  Bruna Moreira <bruna.moreira@indt.org.br>
  20  *  Edjard Mota <edjard.mota@indt.org.br>
  21  *  Ilias Biris <ilias.biris@indt.org.br>
  22  *  Mauricio Lin <mauricio.lin@indt.org.br>
  23  *
  24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  25  *
  26  *  A new process specific entry (smaps) included in /proc. It shows the
  27  *  size of rss for each memory area. The maps entry lacks information
  28  *  about physical memory size (rss) for each mapped file, i.e.,
  29  *  rss information for executables and library files.
  30  *  This additional information is useful for any tools that need to know
  31  *  about physical memory consumption for a process specific library.
  32  *
  33  *  Changelog:
  34  *  21-Feb-2005
  35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  36  *  Pud inclusion in the page table walking.
  37  *
  38  *  ChangeLog:
  39  *  10-Mar-2005
  40  *  10LE Instituto Nokia de Tecnologia - INdT:
  41  *  A better way to walks through the page table as suggested by Hugh Dickins.
  42  *
  43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <paul.mundt@nokia.com>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <asm/uaccess.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/time.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/stat.h>
  56 #include <linux/task_io_accounting_ops.h>
  57 #include <linux/init.h>
  58 #include <linux/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/fdtable.h>
  61 #include <linux/string.h>
  62 #include <linux/seq_file.h>
  63 #include <linux/namei.h>
  64 #include <linux/mnt_namespace.h>
  65 #include <linux/mm.h>
  66 #include <linux/swap.h>
  67 #include <linux/rcupdate.h>
  68 #include <linux/kallsyms.h>
  69 #include <linux/stacktrace.h>
  70 #include <linux/resource.h>
  71 #include <linux/module.h>
  72 #include <linux/mount.h>
  73 #include <linux/security.h>
  74 #include <linux/ptrace.h>
  75 #include <linux/tracehook.h>
  76 #include <linux/cgroup.h>
  77 #include <linux/cpuset.h>
  78 #include <linux/audit.h>
  79 #include <linux/poll.h>
  80 #include <linux/nsproxy.h>
  81 #include <linux/oom.h>
  82 #include <linux/elf.h>
  83 #include <linux/pid_namespace.h>
  84 #include <linux/fs_struct.h>
  85 #include <linux/slab.h>
  86 #include "internal.h"
  87
  88 /* NOTE:
  89  *      Implementing inode permission operations in /proc is almost
  90  *      certainly an error.  Permission checks need to happen during
  91  *      each system call not at open time.  The reason is that most of
  92  *      what we wish to check for permissions in /proc varies at runtime.
  93  *
  94  *      The classic example of a problem is opening file descriptors
  95  *      in /proc for a task before it execs a suid executable.
  96  */
  97
  98 struct pid_entry {
  99         char *name;
 100         int len;
 101         mode_t mode;
 102         const struct inode_operations *iop;
 103         const struct file_operations *fop;
 104         union proc_op op;
 105 };
 106
 107 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 108         .name = (NAME),                                 \
 109         .len  = sizeof(NAME) - 1,                       \
 110         .mode = MODE,                                   \
 111         .iop  = IOP,                                    \
 112         .fop  = FOP,                                    \
 113         .op   = OP,                                     \
 114 }
 115
 116 #define DIR(NAME, MODE, iops, fops)     \
 117         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 118 #define LNK(NAME, get_link)                                     \
 119         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 120                 &proc_pid_link_inode_operations, NULL,          \
 121                 { .proc_get_link = get_link } )
 122 #define REG(NAME, MODE, fops)                           \
 123         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 124 #define INF(NAME, MODE, read)                           \
 125         NOD(NAME, (S_IFREG|(MODE)),                     \
 126                 NULL, &proc_info_file_operations,       \
 127                 { .proc_read = read } )
 128 #define ONE(NAME, MODE, show)                           \
 129         NOD(NAME, (S_IFREG|(MODE)),                     \
 130                 NULL, &proc_single_file_operations,     \
 131                 { .proc_show = show } )
 132
 133 /*
 134  * Count the number of hardlinks for the pid_entry table, excluding the .
 135  * and .. links.
 136  */
 137 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 138         unsigned int n)
 139 {
 140         unsigned int i;
 141         unsigned int count;
 142
 143         count = 0;
 144         for (i = 0; i < n; ++i) {
 145                 if (S_ISDIR(entries[i].mode))
 146                         ++count;
 147         }
 148
 149         return count;
 150 }
 151
 152 static int get_task_root(struct task_struct *task, struct path *root)
 153 {
 154         int result = -ENOENT;
 155
 156         task_lock(task);
 157         if (task->fs) {
 158                 get_fs_root(task->fs, root);
 159                 result = 0;
 160         }
 161         task_unlock(task);
 162         return result;
 163 }
 164
 165 static int proc_cwd_link(struct inode *inode, struct path *path)
 166 {
 167         struct task_struct *task = get_proc_task(inode);
 168         int result = -ENOENT;
 169
 170         if (task) {
 171                 task_lock(task);
 172                 if (task->fs) {
 173                         get_fs_pwd(task->fs, path);
 174                         result = 0;
 175                 }
 176                 task_unlock(task);
 177                 put_task_struct(task);
 178         }
 179         return result;
 180 }
 181
 182 static int proc_root_link(struct inode *inode, struct path *path)
 183 {
 184         struct task_struct *task = get_proc_task(inode);
 185         int result = -ENOENT;
 186
 187         if (task) {
 188                 result = get_task_root(task, path);
 189                 put_task_struct(task);
 190         }
 191         return result;
 192 }
 193
 194 /*
 195  * Return zero if current may access user memory in @task, -error if not.
 196  */
 197 static int check_mem_permission(struct task_struct *task)
 198 {
 199         /*
 200          * A task can always look at itself, in case it chooses
 201          * to use system calls instead of load instructions.
 202          */
 203         if (task == current)
 204                 return 0;
 205
 206         /*
 207          * If current is actively ptrace'ing, and would also be
 208          * permitted to freshly attach with ptrace now, permit it.
 209          */
 210         if (task_is_stopped_or_traced(task)) {
 211                 int match;
 212                 rcu_read_lock();
 213                 match = (tracehook_tracer_task(task) == current);
 214                 rcu_read_unlock();
 215                 if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH))
 216                         return 0;
 217         }
 218
 219         /*
 220          * Noone else is allowed.
 221          */
 222         return -EPERM;
 223 }
 224
 225 struct mm_struct *mm_for_maps(struct task_struct *task)
 226 {
 227         struct mm_struct *mm;
 228
 229         if (mutex_lock_killable(&task->signal->cred_guard_mutex))
 230                 return NULL;
 231
 232         mm = get_task_mm(task);
 233         if (mm && mm != current->mm &&
 234                         !ptrace_may_access(task, PTRACE_MODE_READ)) {
 235                 mmput(mm);
 236                 mm = NULL;
 237         }
 238         mutex_unlock(&task->signal->cred_guard_mutex);
 239
 240         return mm;
 241 }
 242
 243 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 244 {
 245         int res = 0;
 246         unsigned int len;
 247         struct mm_struct *mm = get_task_mm(task);
 248         if (!mm)
 249                 goto out;
 250         if (!mm->arg_end)
 251                 goto out_mm;    /* Shh! No looking before we're done */
 252
 253         len = mm->arg_end - mm->arg_start;
 254
 255         if (len > PAGE_SIZE)
 256                 len = PAGE_SIZE;
 257
 258         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 259
 260         // If the nul at the end of args has been overwritten, then
 261         // assume application is using setproctitle(3).
 262         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 263                 len = strnlen(buffer, res);
 264                 if (len < res) {
 265                     res = len;
 266                 } else {
 267                         len = mm->env_end - mm->env_start;
 268                         if (len > PAGE_SIZE - res)
 269                                 len = PAGE_SIZE - res;
 270                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 271                         res = strnlen(buffer, res);
 272                 }
 273         }
 274 out_mm:
 275         mmput(mm);
 276 out:
 277         return res;
 278 }
 279
 280 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 281 {
 282         int res = 0;
 283         struct mm_struct *mm = get_task_mm(task);
 284         if (mm) {
 285                 unsigned int nwords = 0;
 286                 do {
 287                         nwords += 2;
 288                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 289                 res = nwords * sizeof(mm->saved_auxv[0]);
 290                 if (res > PAGE_SIZE)
 291                         res = PAGE_SIZE;
 292                 memcpy(buffer, mm->saved_auxv, res);
 293                 mmput(mm);
 294         }
 295         return res;
 296 }
 297
 298
 299 #ifdef CONFIG_KALLSYMS
 300 /*
 301  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 302  * Returns the resolved symbol.  If that fails, simply return the address.
 303  */
 304 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 305 {
 306         unsigned long wchan;
 307         char symname[KSYM_NAME_LEN];
 308
 309         wchan = get_wchan(task);
 310
 311         if (lookup_symbol_name(wchan, symname) < 0)
 312                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 313                         return 0;
 314                 else
 315                         return sprintf(buffer, "%lu", wchan);
 316         else
 317                 return sprintf(buffer, "%s", symname);
 318 }
 319 #endif /* CONFIG_KALLSYMS */
 320
 321 #ifdef CONFIG_STACKTRACE
 322
 323 #define MAX_STACK_TRACE_DEPTH   64
 324
 325 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 326                           struct pid *pid, struct task_struct *task)
 327 {
 328         struct stack_trace trace;
 329         unsigned long *entries;
 330         int i;
 331
 332         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 333         if (!entries)
 334                 return -ENOMEM;
 335
 336         trace.nr_entries        = 0;
 337         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 338         trace.entries           = entries;
 339         trace.skip              = 0;
 340         save_stack_trace_tsk(task, &trace);
 341
 342         for (i = 0; i < trace.nr_entries; i++) {
 343                 seq_printf(m, "[<%p>] %pS\n",
 344                            (void *)entries[i], (void *)entries[i]);
 345         }
 346         kfree(entries);
 347
 348         return 0;
 349 }
 350 #endif
 351
 352 #ifdef CONFIG_SCHEDSTATS
 353 /*
 354  * Provides /proc/PID/schedstat
 355  */
 356 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 357 {
 358         return sprintf(buffer, "%llu %llu %lu\n",
 359                         (unsigned long long)task->se.sum_exec_runtime,
 360                         (unsigned long long)task->sched_info.run_delay,
 361                         task->sched_info.pcount);
 362 }
 363 #endif
 364
 365 #ifdef CONFIG_LATENCYTOP
 366 static int lstats_show_proc(struct seq_file *m, void *v)
 367 {
 368         int i;
 369         struct inode *inode = m->private;
 370         struct task_struct *task = get_proc_task(inode);
 371
 372         if (!task)
 373                 return -ESRCH;
 374         seq_puts(m, "Latency Top version : v0.1\n");
 375         for (i = 0; i < 32; i++) {
 376                 struct latency_record *lr = &task->latency_record[i];
 377                 if (lr->backtrace[0]) {
 378                         int q;
 379                         seq_printf(m, "%i %li %li",
 380                                    lr->count, lr->time, lr->max);
 381                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 382                                 unsigned long bt = lr->backtrace[q];
 383                                 if (!bt)
 384                                         break;
 385                                 if (bt == ULONG_MAX)
 386                                         break;
 387                                 seq_printf(m, " %ps", (void *)bt);
 388                         }
 389                         seq_putc(m, '\n');
 390                 }
 391
 392         }
 393         put_task_struct(task);
 394         return 0;
 395 }
 396
 397 static int lstats_open(struct inode *inode, struct file *file)
 398 {
 399         return single_open(file, lstats_show_proc, inode);
 400 }
 401
 402 static ssize_t lstats_write(struct file *file, const char __user *buf,
 403                             size_t count, loff_t *offs)
 404 {
 405         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 406
 407         if (!task)
 408                 return -ESRCH;
 409         clear_all_latency_tracing(task);
 410         put_task_struct(task);
 411
 412         return count;
 413 }
 414
 415 static const struct file_operations proc_lstats_operations = {
 416         .open           = lstats_open,
 417         .read           = seq_read,
 418         .write          = lstats_write,
 419         .llseek         = seq_lseek,
 420         .release        = single_release,
 421 };
 422
 423 #endif
 424
 425 static int proc_oom_score(struct task_struct *task, char *buffer)
 426 {
 427         unsigned long points = 0;
 428
 429         read_lock(&tasklist_lock);
 430         if (pid_alive(task))
 431                 points = oom_badness(task, NULL, NULL,
 432                                         totalram_pages + total_swap_pages);
 433         read_unlock(&tasklist_lock);
 434         return sprintf(buffer, "%lu\n", points);
 435 }
 436
 437 struct limit_names {
 438         char *name;
 439         char *unit;
 440 };
 441
 442 static const struct limit_names lnames[RLIM_NLIMITS] = {
 443         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 444         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 445         [RLIMIT_DATA] = {"Max data size", "bytes"},
 446         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 447         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 448         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 449         [RLIMIT_NPROC] = {"Max processes", "processes"},
 450         [RLIMIT_NOFILE] = {"Max open files", "files"},
 451         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 452         [RLIMIT_AS] = {"Max address space", "bytes"},
 453         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 454         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 455         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 456         [RLIMIT_NICE] = {"Max nice priority", NULL},
 457         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 458         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 459 };
 460
 461 /* Display limits for a process */
 462 static int proc_pid_limits(struct task_struct *task, char *buffer)
 463 {
 464         unsigned int i;
 465         int count = 0;
 466         unsigned long flags;
 467         char *bufptr = buffer;
 468
 469         struct rlimit rlim[RLIM_NLIMITS];
 470
 471         if (!lock_task_sighand(task, &flags))
 472                 return 0;
 473         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 474         unlock_task_sighand(task, &flags);
 475
 476         /*
 477          * print the file header
 478          */
 479         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 480                         "Limit", "Soft Limit", "Hard Limit", "Units");
 481
 482         for (i = 0; i < RLIM_NLIMITS; i++) {
 483                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 484                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 485                                          lnames[i].name, "unlimited");
 486                 else
 487                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 488                                          lnames[i].name, rlim[i].rlim_cur);
 489
 490                 if (rlim[i].rlim_max == RLIM_INFINITY)
 491                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 492                 else
 493                         count += sprintf(&bufptr[count], "%-20lu ",
 494                                          rlim[i].rlim_max);
 495
 496                 if (lnames[i].unit)
 497                         count += sprintf(&bufptr[count], "%-10s\n",
 498                                          lnames[i].unit);
 499                 else
 500                         count += sprintf(&bufptr[count], "\n");
 501         }
 502
 503         return count;
 504 }
 505
 506 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 507 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 508 {
 509         long nr;
 510         unsigned long args[6], sp, pc;
 511
 512         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 513                 return sprintf(buffer, "running\n");
 514
 515         if (nr < 0)
 516                 return sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 517
 518         return sprintf(buffer,
 519                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 520                        nr,
 521                        args[0], args[1], args[2], args[3], args[4], args[5],
 522                        sp, pc);
 523 }
 524 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 525
 526 /************************************************************************/
 527 /*                       Here the fs part begins                        */
 528 /************************************************************************/
 529
 530 /* permission checks */
 531 static int proc_fd_access_allowed(struct inode *inode)
 532 {
 533         struct task_struct *task;
 534         int allowed = 0;
 535         /* Allow access to a task's file descriptors if it is us or we
 536          * may use ptrace attach to the process and find out that
 537          * information.
 538          */
 539         task = get_proc_task(inode);
 540         if (task) {
 541                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 542                 put_task_struct(task);
 543         }
 544         return allowed;
 545 }
 546
 547 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
 548 {
 549         int error;
 550         struct inode *inode = dentry->d_inode;
 551
 552         if (attr->ia_valid & ATTR_MODE)
 553                 return -EPERM;
 554
 555         error = inode_change_ok(inode, attr);
 556         if (error)
 557                 return error;
 558
 559         if ((attr->ia_valid & ATTR_SIZE) &&
 560             attr->ia_size != i_size_read(inode)) {
 561                 error = vmtruncate(inode, attr->ia_size);
 562                 if (error)
 563                         return error;
 564         }
 565
 566         setattr_copy(inode, attr);
 567         mark_inode_dirty(inode);
 568         return 0;
 569 }
 570
 571 static const struct inode_operations proc_def_inode_operations = {
 572         .setattr        = proc_setattr,
 573 };
 574
 575 static int mounts_open_common(struct inode *inode, struct file *file,
 576                               const struct seq_operations *op)
 577 {
 578         struct task_struct *task = get_proc_task(inode);
 579         struct nsproxy *nsp;
 580         struct mnt_namespace *ns = NULL;
 581         struct path root;
 582         struct proc_mounts *p;
 583         int ret = -EINVAL;
 584
 585         if (task) {
 586                 rcu_read_lock();
 587                 nsp = task_nsproxy(task);
 588                 if (nsp) {
 589                         ns = nsp->mnt_ns;
 590                         if (ns)
 591                                 get_mnt_ns(ns);
 592                 }
 593                 rcu_read_unlock();
 594                 if (ns && get_task_root(task, &root) == 0)
 595                         ret = 0;
 596                 put_task_struct(task);
 597         }
 598
 599         if (!ns)
 600                 goto err;
 601         if (ret)
 602                 goto err_put_ns;
 603
 604         ret = -ENOMEM;
 605         p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 606         if (!p)
 607                 goto err_put_path;
 608
 609         file->private_data = &p->m;
 610         ret = seq_open(file, op);
 611         if (ret)
 612                 goto err_free;
 613
 614         p->m.private = p;
 615         p->ns = ns;
 616         p->root = root;
 617         p->event = ns->event;
 618
 619         return 0;
 620
 621  err_free:
 622         kfree(p);
 623  err_put_path:
 624         path_put(&root);
 625  err_put_ns:
 626         put_mnt_ns(ns);
 627  err:
 628         return ret;
 629 }
 630
 631 static int mounts_release(struct inode *inode, struct file *file)
 632 {
 633         struct proc_mounts *p = file->private_data;
 634         path_put(&p->root);
 635         put_mnt_ns(p->ns);
 636         return seq_release(inode, file);
 637 }
 638
 639 static unsigned mounts_poll(struct file *file, poll_table *wait)
 640 {
 641         struct proc_mounts *p = file->private_data;
 642         unsigned res = POLLIN | POLLRDNORM;
 643
 644         poll_wait(file, &p->ns->poll, wait);
 645         if (mnt_had_events(p))
 646                 res |= POLLERR | POLLPRI;
 647
 648         return res;
 649 }
 650
 651 static int mounts_open(struct inode *inode, struct file *file)
 652 {
 653         return mounts_open_common(inode, file, &mounts_op);
 654 }
 655
 656 static const struct file_operations proc_mounts_operations = {
 657         .open           = mounts_open,
 658         .read           = seq_read,
 659         .llseek         = seq_lseek,
 660         .release        = mounts_release,
 661         .poll           = mounts_poll,
 662 };
 663
 664 static int mountinfo_open(struct inode *inode, struct file *file)
 665 {
 666         return mounts_open_common(inode, file, &mountinfo_op);
 667 }
 668
 669 static const struct file_operations proc_mountinfo_operations = {
 670         .open           = mountinfo_open,
 671         .read           = seq_read,
 672         .llseek         = seq_lseek,
 673         .release        = mounts_release,
 674         .poll           = mounts_poll,
 675 };
 676
 677 static int mountstats_open(struct inode *inode, struct file *file)
 678 {
 679         return mounts_open_common(inode, file, &mountstats_op);
 680 }
 681
 682 static const struct file_operations proc_mountstats_operations = {
 683         .open           = mountstats_open,
 684         .read           = seq_read,
 685         .llseek         = seq_lseek,
 686         .release        = mounts_release,
 687 };
 688
 689 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 690
 691 static ssize_t proc_info_read(struct file * file, char __user * buf,
 692                           size_t count, loff_t *ppos)
 693 {
 694         struct inode * inode = file->f_path.dentry->d_inode;
 695         unsigned long page;
 696         ssize_t length;
 697         struct task_struct *task = get_proc_task(inode);
 698
 699         length = -ESRCH;
 700         if (!task)
 701                 goto out_no_task;
 702
 703         if (count > PROC_BLOCK_SIZE)
 704                 count = PROC_BLOCK_SIZE;
 705
 706         length = -ENOMEM;
 707         if (!(page = __get_free_page(GFP_TEMPORARY)))
 708                 goto out;
 709
 710         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 711
 712         if (length >= 0)
 713                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 714         free_page(page);
 715 out:
 716         put_task_struct(task);
 717 out_no_task:
 718         return length;
 719 }
 720
 721 static const struct file_operations proc_info_file_operations = {
 722         .read           = proc_info_read,
 723         .llseek         = generic_file_llseek,
 724 };
 725
 726 static int proc_single_show(struct seq_file *m, void *v)
 727 {
 728         struct inode *inode = m->private;
 729         struct pid_namespace *ns;
 730         struct pid *pid;
 731         struct task_struct *task;
 732         int ret;
 733
 734         ns = inode->i_sb->s_fs_info;
 735         pid = proc_pid(inode);
 736         task = get_pid_task(pid, PIDTYPE_PID);
 737         if (!task)
 738                 return -ESRCH;
 739
 740         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 741
 742         put_task_struct(task);
 743         return ret;
 744 }
 745
 746 static int proc_single_open(struct inode *inode, struct file *filp)
 747 {
 748         int ret;
 749         ret = single_open(filp, proc_single_show, NULL);
 750         if (!ret) {
 751                 struct seq_file *m = filp->private_data;
 752
 753                 m->private = inode;
 754         }
 755         return ret;
 756 }
 757
 758 static const struct file_operations proc_single_file_operations = {
 759         .open           = proc_single_open,
 760         .read           = seq_read,
 761         .llseek         = seq_lseek,
 762         .release        = single_release,
 763 };
 764
 765 static int mem_open(struct inode* inode, struct file* file)
 766 {
 767         file->private_data = (void*)((long)current->self_exec_id);
 768         /* OK to pass negative loff_t, we can catch out-of-range */
 769         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 770         return 0;
 771 }
 772
 773 static ssize_t mem_read(struct file * file, char __user * buf,
 774                         size_t count, loff_t *ppos)
 775 {
 776         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 777         char *page;
 778         unsigned long src = *ppos;
 779         int ret = -ESRCH;
 780         struct mm_struct *mm;
 781
 782         if (!task)
 783                 goto out_no_task;
 784
 785         if (check_mem_permission(task))
 786                 goto out;
 787
 788         ret = -ENOMEM;
 789         page = (char *)__get_free_page(GFP_TEMPORARY);
 790         if (!page)
 791                 goto out;
 792
 793         ret = 0;
 794
 795         mm = get_task_mm(task);
 796         if (!mm)
 797                 goto out_free;
 798
 799         ret = -EIO;
 800
 801         if (file->private_data != (void*)((long)current->self_exec_id))
 802                 goto out_put;
 803
 804         ret = 0;
 805
 806         while (count > 0) {
 807                 int this_len, retval;
 808
 809                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 810                 retval = access_process_vm(task, src, page, this_len, 0);
 811                 if (!retval || check_mem_permission(task)) {
 812                         if (!ret)
 813                                 ret = -EIO;
 814                         break;
 815                 }
 816
 817                 if (copy_to_user(buf, page, retval)) {
 818                         ret = -EFAULT;
 819                         break;
 820                 }
 821
 822                 ret += retval;
 823                 src += retval;
 824                 buf += retval;
 825                 count -= retval;
 826         }
 827         *ppos = src;
 828
 829 out_put:
 830         mmput(mm);
 831 out_free:
 832         free_page((unsigned long) page);
 833 out:
 834         put_task_struct(task);
 835 out_no_task:
 836         return ret;
 837 }
 838
 839 #define mem_write NULL
 840
 841 #ifndef mem_write
 842 /* This is a security hazard */
 843 static ssize_t mem_write(struct file * file, const char __user *buf,
 844                          size_t count, loff_t *ppos)
 845 {
 846         int copied;
 847         char *page;
 848         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 849         unsigned long dst = *ppos;
 850
 851         copied = -ESRCH;
 852         if (!task)
 853                 goto out_no_task;
 854
 855         if (check_mem_permission(task))
 856                 goto out;
 857
 858         copied = -ENOMEM;
 859         page = (char *)__get_free_page(GFP_TEMPORARY);
 860         if (!page)
 861                 goto out;
 862
 863         copied = 0;
 864         while (count > 0) {
 865                 int this_len, retval;
 866
 867                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 868                 if (copy_from_user(page, buf, this_len)) {
 869                         copied = -EFAULT;
 870                         break;
 871                 }
 872                 retval = access_process_vm(task, dst, page, this_len, 1);
 873                 if (!retval) {
 874                         if (!copied)
 875                                 copied = -EIO;
 876                         break;
 877                 }
 878                 copied += retval;
 879                 buf += retval;
 880                 dst += retval;
 881                 count -= retval;
 882         }
 883         *ppos = dst;
 884         free_page((unsigned long) page);
 885 out:
 886         put_task_struct(task);
 887 out_no_task:
 888         return copied;
 889 }
 890 #endif
 891
 892 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 893 {
 894         switch (orig) {
 895         case 0:
 896                 file->f_pos = offset;
 897                 break;
 898         case 1:
 899                 file->f_pos += offset;
 900                 break;
 901         default:
 902                 return -EINVAL;
 903         }
 904         force_successful_syscall_return();
 905         return file->f_pos;
 906 }
 907
 908 static const struct file_operations proc_mem_operations = {
 909         .llseek         = mem_lseek,
 910         .read           = mem_read,
 911         .write          = mem_write,
 912         .open           = mem_open,
 913 };
 914
 915 static ssize_t environ_read(struct file *file, char __user *buf,
 916                         size_t count, loff_t *ppos)
 917 {
 918         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 919         char *page;
 920         unsigned long src = *ppos;
 921         int ret = -ESRCH;
 922         struct mm_struct *mm;
 923
 924         if (!task)
 925                 goto out_no_task;
 926
 927         if (!ptrace_may_access(task, PTRACE_MODE_READ))
 928                 goto out;
 929
 930         ret = -ENOMEM;
 931         page = (char *)__get_free_page(GFP_TEMPORARY);
 932         if (!page)
 933                 goto out;
 934
 935         ret = 0;
 936
 937         mm = get_task_mm(task);
 938         if (!mm)
 939                 goto out_free;
 940
 941         while (count > 0) {
 942                 int this_len, retval, max_len;
 943
 944                 this_len = mm->env_end - (mm->env_start + src);
 945
 946                 if (this_len <= 0)
 947                         break;
 948
 949                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 950                 this_len = (this_len > max_len) ? max_len : this_len;
 951
 952                 retval = access_process_vm(task, (mm->env_start + src),
 953                         page, this_len, 0);
 954
 955                 if (retval <= 0) {
 956                         ret = retval;
 957                         break;
 958                 }
 959
 960                 if (copy_to_user(buf, page, retval)) {
 961                         ret = -EFAULT;
 962                         break;
 963                 }
 964
 965                 ret += retval;
 966                 src += retval;
 967                 buf += retval;
 968                 count -= retval;
 969         }
 970         *ppos = src;
 971
 972         mmput(mm);
 973 out_free:
 974         free_page((unsigned long) page);
 975 out:
 976         put_task_struct(task);
 977 out_no_task:
 978         return ret;
 979 }
 980
 981 static const struct file_operations proc_environ_operations = {
 982         .read           = environ_read,
 983         .llseek         = generic_file_llseek,
 984 };
 985
 986 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 987                                 size_t count, loff_t *ppos)
 988 {
 989         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 990         char buffer[PROC_NUMBUF];
 991         size_t len;
 992         int oom_adjust = OOM_DISABLE;
 993         unsigned long flags;
 994
 995         if (!task)
 996                 return -ESRCH;
 997
 998         if (lock_task_sighand(task, &flags)) {
 999                 oom_adjust = task->signal->oom_adj;
1000                 unlock_task_sighand(task, &flags);
1001         }
1002
1003         put_task_struct(task);
1004
1005         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
1006
1007         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1008 }
1009
1010 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1011                                 size_t count, loff_t *ppos)
1012 {
1013         struct task_struct *task;
1014         char buffer[PROC_NUMBUF];
1015         long oom_adjust;
1016         unsigned long flags;
1017         int err;
1018
1019         memset(buffer, 0, sizeof(buffer));
1020         if (count > sizeof(buffer) - 1)
1021                 count = sizeof(buffer) - 1;
1022         if (copy_from_user(buffer, buf, count)) {
1023                 err = -EFAULT;
1024                 goto out;
1025         }
1026
1027         err = strict_strtol(strstrip(buffer), 0, &oom_adjust);
1028         if (err)
1029                 goto out;
1030         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1031              oom_adjust != OOM_DISABLE) {
1032                 err = -EINVAL;
1033                 goto out;
1034         }
1035
1036         task = get_proc_task(file->f_path.dentry->d_inode);
1037         if (!task) {
1038                 err = -ESRCH;
1039                 goto out;
1040         }
1041
1042         task_lock(task);
1043         if (!task->mm) {
1044                 err = -EINVAL;
1045                 goto err_task_lock;
1046         }
1047
1048         if (!lock_task_sighand(task, &flags)) {
1049                 err = -ESRCH;
1050                 goto err_task_lock;
1051         }
1052
1053         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
1054                 err = -EACCES;
1055                 goto err_sighand;
1056         }
1057
1058         if (oom_adjust != task->signal->oom_adj) {
1059                 if (oom_adjust == OOM_DISABLE)
1060                         atomic_inc(&task->mm->oom_disable_count);
1061                 if (task->signal->oom_adj == OOM_DISABLE)
1062                         atomic_dec(&task->mm->oom_disable_count);
1063         }
1064
1065         /*
1066          * Warn that /proc/pid/oom_adj is deprecated, see
1067          * Documentation/feature-removal-schedule.txt.
1068          */
1069         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, "
1070                         "please use /proc/%d/oom_score_adj instead.\n",
1071                         current->comm, task_pid_nr(current),
1072                         task_pid_nr(task), task_pid_nr(task));
1073         task->signal->oom_adj = oom_adjust;
1074         /*
1075          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1076          * value is always attainable.
1077          */
1078         if (task->signal->oom_adj == OOM_ADJUST_MAX)
1079                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
1080         else
1081                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
1082                                                                 -OOM_DISABLE;
1083 err_sighand:
1084         unlock_task_sighand(task, &flags);
1085 err_task_lock:
1086         task_unlock(task);
1087         put_task_struct(task);
1088 out:
1089         return err < 0 ? err : count;
1090 }
1091
1092 static const struct file_operations proc_oom_adjust_operations = {
1093         .read           = oom_adjust_read,
1094         .write          = oom_adjust_write,
1095         .llseek         = generic_file_llseek,
1096 };
1097
1098 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1099                                         size_t count, loff_t *ppos)
1100 {
1101         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1102         char buffer[PROC_NUMBUF];
1103         int oom_score_adj = OOM_SCORE_ADJ_MIN;
1104         unsigned long flags;
1105         size_t len;
1106
1107         if (!task)
1108                 return -ESRCH;
1109         if (lock_task_sighand(task, &flags)) {
1110                 oom_score_adj = task->signal->oom_score_adj;
1111                 unlock_task_sighand(task, &flags);
1112         }
1113         put_task_struct(task);
1114         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
1115         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1116 }
1117
1118 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1119                                         size_t count, loff_t *ppos)
1120 {
1121         struct task_struct *task;
1122         char buffer[PROC_NUMBUF];
1123         unsigned long flags;
1124         long oom_score_adj;
1125         int err;
1126
1127         memset(buffer, 0, sizeof(buffer));
1128         if (count > sizeof(buffer) - 1)
1129                 count = sizeof(buffer) - 1;
1130         if (copy_from_user(buffer, buf, count)) {
1131                 err = -EFAULT;
1132                 goto out;
1133         }
1134
1135         err = strict_strtol(strstrip(buffer), 0, &oom_score_adj);
1136         if (err)
1137                 goto out;
1138         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1139                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1140                 err = -EINVAL;
1141                 goto out;
1142         }
1143
1144         task = get_proc_task(file->f_path.dentry->d_inode);
1145         if (!task) {
1146                 err = -ESRCH;
1147                 goto out;
1148         }
1149
1150         task_lock(task);
1151         if (!task->mm) {
1152                 err = -EINVAL;
1153                 goto err_task_lock;
1154         }
1155
1156         if (!lock_task_sighand(task, &flags)) {
1157                 err = -ESRCH;
1158                 goto err_task_lock;
1159         }
1160
1161         if (oom_score_adj < task->signal->oom_score_adj &&
1162                         !capable(CAP_SYS_RESOURCE)) {
1163                 err = -EACCES;
1164                 goto err_sighand;
1165         }
1166
1167         if (oom_score_adj != task->signal->oom_score_adj) {
1168                 if (oom_score_adj == OOM_SCORE_ADJ_MIN)
1169                         atomic_inc(&task->mm->oom_disable_count);
1170                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1171                         atomic_dec(&task->mm->oom_disable_count);
1172         }
1173         task->signal->oom_score_adj = oom_score_adj;
1174         /*
1175          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1176          * always attainable.
1177          */
1178         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1179                 task->signal->oom_adj = OOM_DISABLE;
1180         else
1181                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1182                                                         OOM_SCORE_ADJ_MAX;
1183 err_sighand:
1184         unlock_task_sighand(task, &flags);
1185 err_task_lock:
1186         task_unlock(task);
1187         put_task_struct(task);
1188 out:
1189         return err < 0 ? err : count;
1190 }
1191
1192 static const struct file_operations proc_oom_score_adj_operations = {
1193         .read           = oom_score_adj_read,
1194         .write          = oom_score_adj_write,
1195         .llseek         = default_llseek,
1196 };
1197
1198 #ifdef CONFIG_AUDITSYSCALL
1199 #define TMPBUFLEN 21
1200 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1201                                   size_t count, loff_t *ppos)
1202 {
1203         struct inode * inode = file->f_path.dentry->d_inode;
1204         struct task_struct *task = get_proc_task(inode);
1205         ssize_t length;
1206         char tmpbuf[TMPBUFLEN];
1207
1208         if (!task)
1209                 return -ESRCH;
1210         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1211                                 audit_get_loginuid(task));
1212         put_task_struct(task);
1213         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1214 }
1215
1216 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1217                                    size_t count, loff_t *ppos)
1218 {
1219         struct inode * inode = file->f_path.dentry->d_inode;
1220         char *page, *tmp;
1221         ssize_t length;
1222         uid_t loginuid;
1223
1224         if (!capable(CAP_AUDIT_CONTROL))
1225                 return -EPERM;
1226
1227         rcu_read_lock();
1228         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1229                 rcu_read_unlock();
1230                 return -EPERM;
1231         }
1232         rcu_read_unlock();
1233
1234         if (count >= PAGE_SIZE)
1235                 count = PAGE_SIZE - 1;
1236
1237         if (*ppos != 0) {
1238                 /* No partial writes. */
1239                 return -EINVAL;
1240         }
1241         page = (char*)__get_free_page(GFP_TEMPORARY);
1242         if (!page)
1243                 return -ENOMEM;
1244         length = -EFAULT;
1245         if (copy_from_user(page, buf, count))
1246                 goto out_free_page;
1247
1248         page[count] = '\0';
1249         loginuid = simple_strtoul(page, &tmp, 10);
1250         if (tmp == page) {
1251                 length = -EINVAL;
1252                 goto out_free_page;
1253
1254         }
1255         length = audit_set_loginuid(current, loginuid);
1256         if (likely(length == 0))
1257                 length = count;
1258
1259 out_free_page:
1260         free_page((unsigned long) page);
1261         return length;
1262 }
1263
1264 static const struct file_operations proc_loginuid_operations = {
1265         .read           = proc_loginuid_read,
1266         .write          = proc_loginuid_write,
1267         .llseek         = generic_file_llseek,
1268 };
1269
1270 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1271                                   size_t count, loff_t *ppos)
1272 {
1273         struct inode * inode = file->f_path.dentry->d_inode;
1274         struct task_struct *task = get_proc_task(inode);
1275         ssize_t length;
1276         char tmpbuf[TMPBUFLEN];
1277
1278         if (!task)
1279                 return -ESRCH;
1280         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1281                                 audit_get_sessionid(task));
1282         put_task_struct(task);
1283         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1284 }
1285
1286 static const struct file_operations proc_sessionid_operations = {
1287         .read           = proc_sessionid_read,
1288         .llseek         = generic_file_llseek,
1289 };
1290 #endif
1291
1292 #ifdef CONFIG_FAULT_INJECTION
1293 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1294                                       size_t count, loff_t *ppos)
1295 {
1296         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1297         char buffer[PROC_NUMBUF];
1298         size_t len;
1299         int make_it_fail;
1300
1301         if (!task)
1302                 return -ESRCH;
1303         make_it_fail = task->make_it_fail;
1304         put_task_struct(task);
1305
1306         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1307
1308         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1309 }
1310
1311 static ssize_t proc_fault_inject_write(struct file * file,
1312                         const char __user * buf, size_t count, loff_t *ppos)
1313 {
1314         struct task_struct *task;
1315         char buffer[PROC_NUMBUF], *end;
1316         int make_it_fail;
1317
1318         if (!capable(CAP_SYS_RESOURCE))
1319                 return -EPERM;
1320         memset(buffer, 0, sizeof(buffer));
1321         if (count > sizeof(buffer) - 1)
1322                 count = sizeof(buffer) - 1;
1323         if (copy_from_user(buffer, buf, count))
1324                 return -EFAULT;
1325         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1326         if (*end)
1327                 return -EINVAL;
1328         task = get_proc_task(file->f_dentry->d_inode);
1329         if (!task)
1330                 return -ESRCH;
1331         task->make_it_fail = make_it_fail;
1332         put_task_struct(task);
1333
1334         return count;
1335 }
1336
1337 static const struct file_operations proc_fault_inject_operations = {
1338         .read           = proc_fault_inject_read,
1339         .write          = proc_fault_inject_write,
1340         .llseek         = generic_file_llseek,
1341 };
1342 #endif
1343
1344
1345 #ifdef CONFIG_SCHED_DEBUG
1346 /*
1347  * Print out various scheduling related per-task fields:
1348  */
1349 static int sched_show(struct seq_file *m, void *v)
1350 {
1351         struct inode *inode = m->private;
1352         struct task_struct *p;
1353
1354         p = get_proc_task(inode);
1355         if (!p)
1356                 return -ESRCH;
1357         proc_sched_show_task(p, m);
1358
1359         put_task_struct(p);
1360
1361         return 0;
1362 }
1363
1364 static ssize_t
1365 sched_write(struct file *file, const char __user *buf,
1366             size_t count, loff_t *offset)
1367 {
1368         struct inode *inode = file->f_path.dentry->d_inode;
1369         struct task_struct *p;
1370
1371         p = get_proc_task(inode);
1372         if (!p)
1373                 return -ESRCH;
1374         proc_sched_set_task(p);
1375
1376         put_task_struct(p);
1377
1378         return count;
1379 }
1380
1381 static int sched_open(struct inode *inode, struct file *filp)
1382 {
1383         int ret;
1384
1385         ret = single_open(filp, sched_show, NULL);
1386         if (!ret) {
1387                 struct seq_file *m = filp->private_data;
1388
1389                 m->private = inode;
1390         }
1391         return ret;
1392 }
1393
1394 static const struct file_operations proc_pid_sched_operations = {
1395         .open           = sched_open,
1396         .read           = seq_read,
1397         .write          = sched_write,
1398         .llseek         = seq_lseek,
1399         .release        = single_release,
1400 };
1401
1402 #endif
1403
1404 #ifdef CONFIG_SCHED_AUTOGROUP
1405 /*
1406  * Print out autogroup related information:
1407  */
1408 static int sched_autogroup_show(struct seq_file *m, void *v)
1409 {
1410         struct inode *inode = m->private;
1411         struct task_struct *p;
1412
1413         p = get_proc_task(inode);
1414         if (!p)
1415                 return -ESRCH;
1416         proc_sched_autogroup_show_task(p, m);
1417
1418         put_task_struct(p);
1419
1420         return 0;
1421 }
1422
1423 static ssize_t
1424 sched_autogroup_write(struct file *file, const char __user *buf,
1425             size_t count, loff_t *offset)
1426 {
1427         struct inode *inode = file->f_path.dentry->d_inode;
1428         struct task_struct *p;
1429         char buffer[PROC_NUMBUF];
1430         long nice;
1431         int err;
1432
1433         memset(buffer, 0, sizeof(buffer));
1434         if (count > sizeof(buffer) - 1)
1435                 count = sizeof(buffer) - 1;
1436         if (copy_from_user(buffer, buf, count))
1437                 return -EFAULT;
1438
1439         err = strict_strtol(strstrip(buffer), 0, &nice);
1440         if (err)
1441                 return -EINVAL;
1442
1443         p = get_proc_task(inode);
1444         if (!p)
1445                 return -ESRCH;
1446
1447         err = nice;
1448         err = proc_sched_autogroup_set_nice(p, &err);
1449         if (err)
1450                 count = err;
1451
1452         put_task_struct(p);
1453
1454         return count;
1455 }
1456
1457 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1458 {
1459         int ret;
1460
1461         ret = single_open(filp, sched_autogroup_show, NULL);
1462         if (!ret) {
1463                 struct seq_file *m = filp->private_data;
1464
1465                 m->private = inode;
1466         }
1467         return ret;
1468 }
1469
1470 static const struct file_operations proc_pid_sched_autogroup_operations = {
1471         .open           = sched_autogroup_open,
1472         .read           = seq_read,
1473         .write          = sched_autogroup_write,
1474         .llseek         = seq_lseek,
1475         .release        = single_release,
1476 };
1477
1478 #endif /* CONFIG_SCHED_AUTOGROUP */
1479
1480 static ssize_t comm_write(struct file *file, const char __user *buf,
1481                                 size_t count, loff_t *offset)
1482 {
1483         struct inode *inode = file->f_path.dentry->d_inode;
1484         struct task_struct *p;
1485         char buffer[TASK_COMM_LEN];
1486
1487         memset(buffer, 0, sizeof(buffer));
1488         if (count > sizeof(buffer) - 1)
1489                 count = sizeof(buffer) - 1;
1490         if (copy_from_user(buffer, buf, count))
1491                 return -EFAULT;
1492
1493         p = get_proc_task(inode);
1494         if (!p)
1495                 return -ESRCH;
1496
1497         if (same_thread_group(current, p))
1498                 set_task_comm(p, buffer);
1499         else
1500                 count = -EINVAL;
1501
1502         put_task_struct(p);
1503
1504         return count;
1505 }
1506
1507 static int comm_show(struct seq_file *m, void *v)
1508 {
1509         struct inode *inode = m->private;
1510         struct task_struct *p;
1511
1512         p = get_proc_task(inode);
1513         if (!p)
1514                 return -ESRCH;
1515
1516         task_lock(p);
1517         seq_printf(m, "%s\n", p->comm);
1518         task_unlock(p);
1519
1520         put_task_struct(p);
1521
1522         return 0;
1523 }
1524
1525 static int comm_open(struct inode *inode, struct file *filp)
1526 {
1527         int ret;
1528
1529         ret = single_open(filp, comm_show, NULL);
1530         if (!ret) {
1531                 struct seq_file *m = filp->private_data;
1532
1533                 m->private = inode;
1534         }
1535         return ret;
1536 }
1537
1538 static const struct file_operations proc_pid_set_comm_operations = {
1539         .open           = comm_open,
1540         .read           = seq_read,
1541         .write          = comm_write,
1542         .llseek         = seq_lseek,
1543         .release        = single_release,
1544 };
1545
1546 /*
1547  * We added or removed a vma mapping the executable. The vmas are only mapped
1548  * during exec and are not mapped with the mmap system call.
1549  * Callers must hold down_write() on the mm's mmap_sem for these
1550  */
1551 void added_exe_file_vma(struct mm_struct *mm)
1552 {
1553         mm->num_exe_file_vmas++;
1554 }
1555
1556 void removed_exe_file_vma(struct mm_struct *mm)
1557 {
1558         mm->num_exe_file_vmas--;
1559         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1560                 fput(mm->exe_file);
1561                 mm->exe_file = NULL;
1562         }
1563
1564 }
1565
1566 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1567 {
1568         if (new_exe_file)
1569                 get_file(new_exe_file);
1570         if (mm->exe_file)
1571                 fput(mm->exe_file);
1572         mm->exe_file = new_exe_file;
1573         mm->num_exe_file_vmas = 0;
1574 }
1575
1576 struct file *get_mm_exe_file(struct mm_struct *mm)
1577 {
1578         struct file *exe_file;
1579
1580         /* We need mmap_sem to protect against races with removal of
1581          * VM_EXECUTABLE vmas */
1582         down_read(&mm->mmap_sem);
1583         exe_file = mm->exe_file;
1584         if (exe_file)
1585                 get_file(exe_file);
1586         up_read(&mm->mmap_sem);
1587         return exe_file;
1588 }
1589
1590 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1591 {
1592         /* It's safe to write the exe_file pointer without exe_file_lock because
1593          * this is called during fork when the task is not yet in /proc */
1594         newmm->exe_file = get_mm_exe_file(oldmm);
1595 }
1596
1597 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1598 {
1599         struct task_struct *task;
1600         struct mm_struct *mm;
1601         struct file *exe_file;
1602
1603         task = get_proc_task(inode);
1604         if (!task)
1605                 return -ENOENT;
1606         mm = get_task_mm(task);
1607         put_task_struct(task);
1608         if (!mm)
1609                 return -ENOENT;
1610         exe_file = get_mm_exe_file(mm);
1611         mmput(mm);
1612         if (exe_file) {
1613                 *exe_path = exe_file->f_path;
1614                 path_get(&exe_file->f_path);
1615                 fput(exe_file);
1616                 return 0;
1617         } else
1618                 return -ENOENT;
1619 }
1620
1621 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1622 {
1623         struct inode *inode = dentry->d_inode;
1624         int error = -EACCES;
1625
1626         /* We don't need a base pointer in the /proc filesystem */
1627         path_put(&nd->path);
1628
1629         /* Are we allowed to snoop on the tasks file descriptors? */
1630         if (!proc_fd_access_allowed(inode))
1631                 goto out;
1632
1633         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1634 out:
1635         return ERR_PTR(error);
1636 }
1637
1638 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1639 {
1640         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1641         char *pathname;
1642         int len;
1643
1644         if (!tmp)
1645                 return -ENOMEM;
1646
1647         pathname = d_path(path, tmp, PAGE_SIZE);
1648         len = PTR_ERR(pathname);
1649         if (IS_ERR(pathname))
1650                 goto out;
1651         len = tmp + PAGE_SIZE - 1 - pathname;
1652
1653         if (len > buflen)
1654                 len = buflen;
1655         if (copy_to_user(buffer, pathname, len))
1656                 len = -EFAULT;
1657  out:
1658         free_page((unsigned long)tmp);
1659         return len;
1660 }
1661
1662 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1663 {
1664         int error = -EACCES;
1665         struct inode *inode = dentry->d_inode;
1666         struct path path;
1667
1668         /* Are we allowed to snoop on the tasks file descriptors? */
1669         if (!proc_fd_access_allowed(inode))
1670                 goto out;
1671
1672         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1673         if (error)
1674                 goto out;
1675
1676         error = do_proc_readlink(&path, buffer, buflen);
1677         path_put(&path);
1678 out:
1679         return error;
1680 }
1681
1682 static const struct inode_operations proc_pid_link_inode_operations = {
1683         .readlink       = proc_pid_readlink,
1684         .follow_link    = proc_pid_follow_link,
1685         .setattr        = proc_setattr,
1686 };
1687
1688
1689 /* building an inode */
1690
1691 static int task_dumpable(struct task_struct *task)
1692 {
1693         int dumpable = 0;
1694         struct mm_struct *mm;
1695
1696         task_lock(task);
1697         mm = task->mm;
1698         if (mm)
1699                 dumpable = get_dumpable(mm);
1700         task_unlock(task);
1701         if(dumpable == 1)
1702                 return 1;
1703         return 0;
1704 }
1705
1706
1707 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1708 {
1709         struct inode * inode;
1710         struct proc_inode *ei;
1711         const struct cred *cred;
1712
1713         /* We need a new inode */
1714
1715         inode = new_inode(sb);
1716         if (!inode)
1717                 goto out;
1718
1719         /* Common stuff */
1720         ei = PROC_I(inode);
1721         inode->i_ino = get_next_ino();
1722         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1723         inode->i_op = &proc_def_inode_operations;
1724
1725         /*
1726          * grab the reference to task.
1727          */
1728         ei->pid = get_task_pid(task, PIDTYPE_PID);
1729         if (!ei->pid)
1730                 goto out_unlock;
1731
1732         if (task_dumpable(task)) {
1733                 rcu_read_lock();
1734                 cred = __task_cred(task);
1735                 inode->i_uid = cred->euid;
1736                 inode->i_gid = cred->egid;
1737                 rcu_read_unlock();
1738         }
1739         security_task_to_inode(task, inode);
1740
1741 out:
1742         return inode;
1743
1744 out_unlock:
1745         iput(inode);
1746         return NULL;
1747 }
1748
1749 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1750 {
1751         struct inode *inode = dentry->d_inode;
1752         struct task_struct *task;
1753         const struct cred *cred;
1754
1755         generic_fillattr(inode, stat);
1756
1757         rcu_read_lock();
1758         stat->uid = 0;
1759         stat->gid = 0;
1760         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1761         if (task) {
1762                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1763                     task_dumpable(task)) {
1764                         cred = __task_cred(task);
1765                         stat->uid = cred->euid;
1766                         stat->gid = cred->egid;
1767                 }
1768         }
1769         rcu_read_unlock();
1770         return 0;
1771 }
1772
1773 /* dentry stuff */
1774
1775 /*
1776  *      Exceptional case: normally we are not allowed to unhash a busy
1777  * directory. In this case, however, we can do it - no aliasing problems
1778  * due to the way we treat inodes.
1779  *
1780  * Rewrite the inode's ownerships here because the owning task may have
1781  * performed a setuid(), etc.
1782  *
1783  * Before the /proc/pid/status file was created the only way to read
1784  * the effective uid of a /process was to stat /proc/pid.  Reading
1785  * /proc/pid/status is slow enough that procps and other packages
1786  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1787  * made this apply to all per process world readable and executable
1788  * directories.
1789  */
1790 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1791 {
1792         struct inode *inode;
1793         struct task_struct *task;
1794         const struct cred *cred;
1795
1796         if (nd && nd->flags & LOOKUP_RCU)
1797                 return -ECHILD;
1798
1799         inode = dentry->d_inode;
1800         task = get_proc_task(inode);
1801
1802         if (task) {
1803                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1804                     task_dumpable(task)) {
1805                         rcu_read_lock();
1806                         cred = __task_cred(task);
1807                         inode->i_uid = cred->euid;
1808                         inode->i_gid = cred->egid;
1809                         rcu_read_unlock();
1810                 } else {
1811                         inode->i_uid = 0;
1812                         inode->i_gid = 0;
1813                 }
1814                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1815                 security_task_to_inode(task, inode);
1816                 put_task_struct(task);
1817                 return 1;
1818         }
1819         d_drop(dentry);
1820         return 0;
1821 }
1822
1823 static int pid_delete_dentry(const struct dentry * dentry)
1824 {
1825         /* Is the task we represent dead?
1826          * If so, then don't put the dentry on the lru list,
1827          * kill it immediately.
1828          */
1829         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1830 }
1831
1832 static const struct dentry_operations pid_dentry_operations =
1833 {
1834         .d_revalidate   = pid_revalidate,
1835         .d_delete       = pid_delete_dentry,
1836 };
1837
1838 /* Lookups */
1839
1840 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1841                                 struct task_struct *, const void *);
1842
1843 /*
1844  * Fill a directory entry.
1845  *
1846  * If possible create the dcache entry and derive our inode number and
1847  * file type from dcache entry.
1848  *
1849  * Since all of the proc inode numbers are dynamically generated, the inode
1850  * numbers do not exist until the inode is cache.  This means creating the
1851  * the dcache entry in readdir is necessary to keep the inode numbers
1852  * reported by readdir in sync with the inode numbers reported
1853  * by stat.
1854  */
1855 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1856         char *name, int len,
1857         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1858 {
1859         struct dentry *child, *dir = filp->f_path.dentry;
1860         struct inode *inode;
1861         struct qstr qname;
1862         ino_t ino = 0;
1863         unsigned type = DT_UNKNOWN;
1864
1865         qname.name = name;
1866         qname.len  = len;
1867         qname.hash = full_name_hash(name, len);
1868
1869         child = d_lookup(dir, &qname);
1870         if (!child) {
1871                 struct dentry *new;
1872                 new = d_alloc(dir, &qname);
1873                 if (new) {
1874                         child = instantiate(dir->d_inode, new, task, ptr);
1875                         if (child)
1876                                 dput(new);
1877                         else
1878                                 child = new;
1879                 }
1880         }
1881         if (!child || IS_ERR(child) || !child->d_inode)
1882                 goto end_instantiate;
1883         inode = child->d_inode;
1884         if (inode) {
1885                 ino = inode->i_ino;
1886                 type = inode->i_mode >> 12;
1887         }
1888         dput(child);
1889 end_instantiate:
1890         if (!ino)
1891                 ino = find_inode_number(dir, &qname);
1892         if (!ino)
1893                 ino = 1;
1894         return filldir(dirent, name, len, filp->f_pos, ino, type);
1895 }
1896
1897 static unsigned name_to_int(struct dentry *dentry)
1898 {
1899         const char *name = dentry->d_name.name;
1900         int len = dentry->d_name.len;
1901         unsigned n = 0;
1902
1903         if (len > 1 && *name == '0')
1904                 goto out;
1905         while (len-- > 0) {
1906                 unsigned c = *name++ - '0';
1907                 if (c > 9)
1908                         goto out;
1909                 if (n >= (~0U-9)/10)
1910                         goto out;
1911                 n *= 10;
1912                 n += c;
1913         }
1914         return n;
1915 out:
1916         return ~0U;
1917 }
1918
1919 #define PROC_FDINFO_MAX 64
1920
1921 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1922 {
1923         struct task_struct *task = get_proc_task(inode);
1924         struct files_struct *files = NULL;
1925         struct file *file;
1926         int fd = proc_fd(inode);
1927
1928         if (task) {
1929                 files = get_files_struct(task);
1930                 put_task_struct(task);
1931         }
1932         if (files) {
1933                 /*
1934                  * We are not taking a ref to the file structure, so we must
1935                  * hold ->file_lock.
1936                  */
1937                 spin_lock(&files->file_lock);
1938                 file = fcheck_files(files, fd);
1939                 if (file) {
1940                         if (path) {
1941                                 *path = file->f_path;
1942                                 path_get(&file->f_path);
1943                         }
1944                         if (info)
1945                                 snprintf(info, PROC_FDINFO_MAX,
1946                                          "pos:\t%lli\n"
1947                                          "flags:\t0%o\n",
1948                                          (long long) file->f_pos,
1949                                          file->f_flags);
1950                         spin_unlock(&files->file_lock);
1951                         put_files_struct(files);
1952                         return 0;
1953                 }
1954                 spin_unlock(&files->file_lock);
1955                 put_files_struct(files);
1956         }
1957         return -ENOENT;
1958 }
1959
1960 static int proc_fd_link(struct inode *inode, struct path *path)
1961 {
1962         return proc_fd_info(inode, path, NULL);
1963 }
1964
1965 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1966 {
1967         struct inode *inode;
1968         struct task_struct *task;
1969         int fd;
1970         struct files_struct *files;
1971         const struct cred *cred;
1972
1973         if (nd && nd->flags & LOOKUP_RCU)
1974                 return -ECHILD;
1975
1976         inode = dentry->d_inode;
1977         task = get_proc_task(inode);
1978         fd = proc_fd(inode);
1979
1980         if (task) {
1981                 files = get_files_struct(task);
1982                 if (files) {
1983                         rcu_read_lock();
1984                         if (fcheck_files(files, fd)) {
1985                                 rcu_read_unlock();
1986                                 put_files_struct(files);
1987                                 if (task_dumpable(task)) {
1988                                         rcu_read_lock();
1989                                         cred = __task_cred(task);
1990                                         inode->i_uid = cred->euid;
1991                                         inode->i_gid = cred->egid;
1992                                         rcu_read_unlock();
1993                                 } else {
1994                                         inode->i_uid = 0;
1995                                         inode->i_gid = 0;
1996                                 }
1997                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1998                                 security_task_to_inode(task, inode);
1999                                 put_task_struct(task);
2000                                 return 1;
2001                         }
2002                         rcu_read_unlock();
2003                         put_files_struct(files);
2004                 }
2005                 put_task_struct(task);
2006         }
2007         d_drop(dentry);
2008         return 0;
2009 }
2010
2011 static const struct dentry_operations tid_fd_dentry_operations =
2012 {
2013         .d_revalidate   = tid_fd_revalidate,
2014         .d_delete       = pid_delete_dentry,
2015 };
2016
2017 static struct dentry *proc_fd_instantiate(struct inode *dir,
2018         struct dentry *dentry, struct task_struct *task, const void *ptr)
2019 {
2020         unsigned fd = *(const unsigned *)ptr;
2021         struct file *file;
2022         struct files_struct *files;
2023         struct inode *inode;
2024         struct proc_inode *ei;
2025         struct dentry *error = ERR_PTR(-ENOENT);
2026
2027         inode = proc_pid_make_inode(dir->i_sb, task);
2028         if (!inode)
2029                 goto out;
2030         ei = PROC_I(inode);
2031         ei->fd = fd;
2032         files = get_files_struct(task);
2033         if (!files)
2034                 goto out_iput;
2035         inode->i_mode = S_IFLNK;
2036
2037         /*
2038          * We are not taking a ref to the file structure, so we must
2039          * hold ->file_lock.
2040          */
2041         spin_lock(&files->file_lock);
2042         file = fcheck_files(files, fd);
2043         if (!file)
2044                 goto out_unlock;
2045         if (file->f_mode & FMODE_READ)
2046                 inode->i_mode |= S_IRUSR | S_IXUSR;
2047         if (file->f_mode & FMODE_WRITE)
2048                 inode->i_mode |= S_IWUSR | S_IXUSR;
2049         spin_unlock(&files->file_lock);
2050         put_files_struct(files);
2051
2052         inode->i_op = &proc_pid_link_inode_operations;
2053         inode->i_size = 64;
2054         ei->op.proc_get_link = proc_fd_link;
2055         d_set_d_op(dentry, &tid_fd_dentry_operations);
2056         d_add(dentry, inode);
2057         /* Close the race of the process dying before we return the dentry */
2058         if (tid_fd_revalidate(dentry, NULL))
2059                 error = NULL;
2060
2061  out:
2062         return error;
2063 out_unlock:
2064         spin_unlock(&files->file_lock);
2065         put_files_struct(files);
2066 out_iput:
2067         iput(inode);
2068         goto out;
2069 }
2070
2071 static struct dentry *proc_lookupfd_common(struct inode *dir,
2072                                            struct dentry *dentry,
2073                                            instantiate_t instantiate)
2074 {
2075         struct task_struct *task = get_proc_task(dir);
2076         unsigned fd = name_to_int(dentry);
2077         struct dentry *result = ERR_PTR(-ENOENT);
2078
2079         if (!task)
2080                 goto out_no_task;
2081         if (fd == ~0U)
2082                 goto out;
2083
2084         result = instantiate(dir, dentry, task, &fd);
2085 out:
2086         put_task_struct(task);
2087 out_no_task:
2088         return result;
2089 }
2090
2091 static int proc_readfd_common(struct file * filp, void * dirent,
2092                               filldir_t filldir, instantiate_t instantiate)
2093 {
2094         struct dentry *dentry = filp->f_path.dentry;
2095         struct inode *inode = dentry->d_inode;
2096         struct task_struct *p = get_proc_task(inode);
2097         unsigned int fd, ino;
2098         int retval;
2099         struct files_struct * files;
2100
2101         retval = -ENOENT;
2102         if (!p)
2103                 goto out_no_task;
2104         retval = 0;
2105
2106         fd = filp->f_pos;
2107         switch (fd) {
2108                 case 0:
2109                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
2110                                 goto out;
2111                         filp->f_pos++;
2112                 case 1:
2113                         ino = parent_ino(dentry);
2114                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2115                                 goto out;
2116                         filp->f_pos++;
2117                 default:
2118                         files = get_files_struct(p);
2119                         if (!files)
2120                                 goto out;
2121                         rcu_read_lock();
2122                         for (fd = filp->f_pos-2;
2123                              fd < files_fdtable(files)->max_fds;
2124                              fd++, filp->f_pos++) {
2125                                 char name[PROC_NUMBUF];
2126                                 int len;
2127
2128                                 if (!fcheck_files(files, fd))
2129                                         continue;
2130                                 rcu_read_unlock();
2131
2132                                 len = snprintf(name, sizeof(name), "%d", fd);
2133                                 if (proc_fill_cache(filp, dirent, filldir,
2134                                                     name, len, instantiate,
2135                                                     p, &fd) < 0) {
2136                                         rcu_read_lock();
2137                                         break;
2138                                 }
2139                                 rcu_read_lock();
2140                         }
2141                         rcu_read_unlock();
2142                         put_files_struct(files);
2143         }
2144 out:
2145         put_task_struct(p);
2146 out_no_task:
2147         return retval;
2148 }
2149
2150 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
2151                                     struct nameidata *nd)
2152 {
2153         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
2154 }
2155
2156 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
2157 {
2158         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
2159 }
2160
2161 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
2162                                       size_t len, loff_t *ppos)
2163 {
2164         char tmp[PROC_FDINFO_MAX];
2165         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
2166         if (!err)
2167                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
2168         return err;
2169 }
2170
2171 static const struct file_operations proc_fdinfo_file_operations = {
2172         .open           = nonseekable_open,
2173         .read           = proc_fdinfo_read,
2174         .llseek         = no_llseek,
2175 };
2176
2177 static const struct file_operations proc_fd_operations = {
2178         .read           = generic_read_dir,
2179         .readdir        = proc_readfd,
2180         .llseek         = default_llseek,
2181 };
2182
2183 /*
2184  * /proc/pid/fd needs a special permission handler so that a process can still
2185  * access /proc/self/fd after it has executed a setuid().
2186  */
2187 static int proc_fd_permission(struct inode *inode, int mask, unsigned int flags)
2188 {
2189         int rv;
2190
2191         if (flags & IPERM_FLAG_RCU)
2192                 return -ECHILD;
2193         rv = generic_permission(inode, mask, flags, NULL);
2194         if (rv == 0)
2195                 return 0;
2196         if (task_pid(current) == proc_pid(inode))
2197                 rv = 0;
2198         return rv;
2199 }
2200
2201 /*
2202  * proc directories can do almost nothing..
2203  */
2204 static const struct inode_operations proc_fd_inode_operations = {
2205         .lookup         = proc_lookupfd,
2206         .permission     = proc_fd_permission,
2207         .setattr        = proc_setattr,
2208 };
2209
2210 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2211         struct dentry *dentry, struct task_struct *task, const void *ptr)
2212 {
2213         unsigned fd = *(unsigned *)ptr;
2214         struct inode *inode;
2215         struct proc_inode *ei;
2216         struct dentry *error = ERR_PTR(-ENOENT);
2217
2218         inode = proc_pid_make_inode(dir->i_sb, task);
2219         if (!inode)
2220                 goto out;
2221         ei = PROC_I(inode);
2222         ei->fd = fd;
2223         inode->i_mode = S_IFREG | S_IRUSR;
2224         inode->i_fop = &proc_fdinfo_file_operations;
2225         d_set_d_op(dentry, &tid_fd_dentry_operations);
2226         d_add(dentry, inode);
2227         /* Close the race of the process dying before we return the dentry */
2228         if (tid_fd_revalidate(dentry, NULL))
2229                 error = NULL;
2230
2231  out:
2232         return error;
2233 }
2234
2235 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2236                                         struct dentry *dentry,
2237                                         struct nameidata *nd)
2238 {
2239         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2240 }
2241
2242 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2243 {
2244         return proc_readfd_common(filp, dirent, filldir,
2245                                   proc_fdinfo_instantiate);
2246 }
2247
2248 static const struct file_operations proc_fdinfo_operations = {
2249         .read           = generic_read_dir,
2250         .readdir        = proc_readfdinfo,
2251         .llseek         = default_llseek,
2252 };
2253
2254 /*
2255  * proc directories can do almost nothing..
2256  */
2257 static const struct inode_operations proc_fdinfo_inode_operations = {
2258         .lookup         = proc_lookupfdinfo,
2259         .setattr        = proc_setattr,
2260 };
2261
2262
2263 static struct dentry *proc_pident_instantiate(struct inode *dir,
2264         struct dentry *dentry, struct task_struct *task, const void *ptr)
2265 {
2266         const struct pid_entry *p = ptr;
2267         struct inode *inode;
2268         struct proc_inode *ei;
2269         struct dentry *error = ERR_PTR(-ENOENT);
2270
2271         inode = proc_pid_make_inode(dir->i_sb, task);
2272         if (!inode)
2273                 goto out;
2274
2275         ei = PROC_I(inode);
2276         inode->i_mode = p->mode;
2277         if (S_ISDIR(inode->i_mode))
2278                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
2279         if (p->iop)
2280                 inode->i_op = p->iop;
2281         if (p->fop)
2282                 inode->i_fop = p->fop;
2283         ei->op = p->op;
2284         d_set_d_op(dentry, &pid_dentry_operations);
2285         d_add(dentry, inode);
2286         /* Close the race of the process dying before we return the dentry */
2287         if (pid_revalidate(dentry, NULL))
2288                 error = NULL;
2289 out:
2290         return error;
2291 }
2292
2293 static struct dentry *proc_pident_lookup(struct inode *dir,
2294                                          struct dentry *dentry,
2295                                          const struct pid_entry *ents,
2296                                          unsigned int nents)
2297 {
2298         struct dentry *error;
2299         struct task_struct *task = get_proc_task(dir);
2300         const struct pid_entry *p, *last;
2301
2302         error = ERR_PTR(-ENOENT);
2303
2304         if (!task)
2305                 goto out_no_task;
2306
2307         /*
2308          * Yes, it does not scale. And it should not. Don't add
2309          * new entries into /proc/<tgid>/ without very good reasons.
2310          */
2311         last = &ents[nents - 1];
2312         for (p = ents; p <= last; p++) {
2313                 if (p->len != dentry->d_name.len)
2314                         continue;
2315                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2316                         break;
2317         }
2318         if (p > last)
2319                 goto out;
2320
2321         error = proc_pident_instantiate(dir, dentry, task, p);
2322 out:
2323         put_task_struct(task);
2324 out_no_task:
2325         return error;
2326 }
2327
2328 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2329         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2330 {
2331         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2332                                 proc_pident_instantiate, task, p);
2333 }
2334
2335 static int proc_pident_readdir(struct file *filp,
2336                 void *dirent, filldir_t filldir,
2337                 const struct pid_entry *ents, unsigned int nents)
2338 {
2339         int i;
2340         struct dentry *dentry = filp->f_path.dentry;
2341         struct inode *inode = dentry->d_inode;
2342         struct task_struct *task = get_proc_task(inode);
2343         const struct pid_entry *p, *last;
2344         ino_t ino;
2345         int ret;
2346
2347         ret = -ENOENT;
2348         if (!task)
2349                 goto out_no_task;
2350
2351         ret = 0;
2352         i = filp->f_pos;
2353         switch (i) {
2354         case 0:
2355                 ino = inode->i_ino;
2356                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2357                         goto out;
2358                 i++;
2359                 filp->f_pos++;
2360                 /* fall through */
2361         case 1:
2362                 ino = parent_ino(dentry);
2363                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2364                         goto out;
2365                 i++;
2366                 filp->f_pos++;
2367                 /* fall through */
2368         default:
2369                 i -= 2;
2370                 if (i >= nents) {
2371                         ret = 1;
2372                         goto out;
2373                 }
2374                 p = ents + i;
2375                 last = &ents[nents - 1];
2376                 while (p <= last) {
2377                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2378                                 goto out;
2379                         filp->f_pos++;
2380                         p++;
2381                 }
2382         }
2383
2384         ret = 1;
2385 out:
2386         put_task_struct(task);
2387 out_no_task:
2388         return ret;
2389 }
2390
2391 #ifdef CONFIG_SECURITY
2392 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2393                                   size_t count, loff_t *ppos)
2394 {
2395         struct inode * inode = file->f_path.dentry->d_inode;
2396         char *p = NULL;
2397         ssize_t length;
2398         struct task_struct *task = get_proc_task(inode);
2399
2400         if (!task)
2401                 return -ESRCH;
2402
2403         length = security_getprocattr(task,
2404                                       (char*)file->f_path.dentry->d_name.name,
2405                                       &p);
2406         put_task_struct(task);
2407         if (length > 0)
2408                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2409         kfree(p);
2410         return length;
2411 }
2412
2413 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2414                                    size_t count, loff_t *ppos)
2415 {
2416         struct inode * inode = file->f_path.dentry->d_inode;
2417         char *page;
2418         ssize_t length;
2419         struct task_struct *task = get_proc_task(inode);
2420
2421         length = -ESRCH;
2422         if (!task)
2423                 goto out_no_task;
2424         if (count > PAGE_SIZE)
2425                 count = PAGE_SIZE;
2426
2427         /* No partial writes. */
2428         length = -EINVAL;
2429         if (*ppos != 0)
2430                 goto out;
2431
2432         length = -ENOMEM;
2433         page = (char*)__get_free_page(GFP_TEMPORARY);
2434         if (!page)
2435                 goto out;
2436
2437         length = -EFAULT;
2438         if (copy_from_user(page, buf, count))
2439                 goto out_free;
2440
2441         /* Guard against adverse ptrace interaction */
2442         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2443         if (length < 0)
2444                 goto out_free;
2445
2446         length = security_setprocattr(task,
2447                                       (char*)file->f_path.dentry->d_name.name,
2448                                       (void*)page, count);
2449         mutex_unlock(&task->signal->cred_guard_mutex);
2450 out_free:
2451         free_page((unsigned long) page);
2452 out:
2453         put_task_struct(task);
2454 out_no_task:
2455         return length;
2456 }
2457
2458 static const struct file_operations proc_pid_attr_operations = {
2459         .read           = proc_pid_attr_read,
2460         .write          = proc_pid_attr_write,
2461         .llseek         = generic_file_llseek,
2462 };
2463
2464 static const struct pid_entry attr_dir_stuff[] = {
2465         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2466         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2467         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2468         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2469         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2470         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2471 };
2472
2473 static int proc_attr_dir_readdir(struct file * filp,
2474                              void * dirent, filldir_t filldir)
2475 {
2476         return proc_pident_readdir(filp,dirent,filldir,
2477                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2478 }
2479
2480 static const struct file_operations proc_attr_dir_operations = {
2481         .read           = generic_read_dir,
2482         .readdir        = proc_attr_dir_readdir,
2483         .llseek         = default_llseek,
2484 };
2485
2486 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2487                                 struct dentry *dentry, struct nameidata *nd)
2488 {
2489         return proc_pident_lookup(dir, dentry,
2490                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2491 }
2492
2493 static const struct inode_operations proc_attr_dir_inode_operations = {
2494         .lookup         = proc_attr_dir_lookup,
2495         .getattr        = pid_getattr,
2496         .setattr        = proc_setattr,
2497 };
2498
2499 #endif
2500
2501 #ifdef CONFIG_ELF_CORE
2502 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2503                                          size_t count, loff_t *ppos)
2504 {
2505         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2506         struct mm_struct *mm;
2507         char buffer[PROC_NUMBUF];
2508         size_t len;
2509         int ret;
2510
2511         if (!task)
2512                 return -ESRCH;
2513
2514         ret = 0;
2515         mm = get_task_mm(task);
2516         if (mm) {
2517                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2518                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2519                                 MMF_DUMP_FILTER_SHIFT));
2520                 mmput(mm);
2521                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2522         }
2523
2524         put_task_struct(task);
2525
2526         return ret;
2527 }
2528
2529 static ssize_t proc_coredump_filter_write(struct file *file,
2530                                           const char __user *buf,
2531                                           size_t count,
2532                                           loff_t *ppos)
2533 {
2534         struct task_struct *task;
2535         struct mm_struct *mm;
2536         char buffer[PROC_NUMBUF], *end;
2537         unsigned int val;
2538         int ret;
2539         int i;
2540         unsigned long mask;
2541
2542         ret = -EFAULT;
2543         memset(buffer, 0, sizeof(buffer));
2544         if (count > sizeof(buffer) - 1)
2545                 count = sizeof(buffer) - 1;
2546         if (copy_from_user(buffer, buf, count))
2547                 goto out_no_task;
2548
2549         ret = -EINVAL;
2550         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2551         if (*end == '\n')
2552                 end++;
2553         if (end - buffer == 0)
2554                 goto out_no_task;
2555
2556         ret = -ESRCH;
2557         task = get_proc_task(file->f_dentry->d_inode);
2558         if (!task)
2559                 goto out_no_task;
2560
2561         ret = end - buffer;
2562         mm = get_task_mm(task);
2563         if (!mm)
2564                 goto out_no_mm;
2565
2566         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2567                 if (val & mask)
2568                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2569                 else
2570                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2571         }
2572
2573         mmput(mm);
2574  out_no_mm:
2575         put_task_struct(task);
2576  out_no_task:
2577         return ret;
2578 }
2579
2580 static const struct file_operations proc_coredump_filter_operations = {
2581         .read           = proc_coredump_filter_read,
2582         .write          = proc_coredump_filter_write,
2583         .llseek         = generic_file_llseek,
2584 };
2585 #endif
2586
2587 /*
2588  * /proc/self:
2589  */
2590 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2591                               int buflen)
2592 {
2593         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2594         pid_t tgid = task_tgid_nr_ns(current, ns);
2595         char tmp[PROC_NUMBUF];
2596         if (!tgid)
2597                 return -ENOENT;
2598         sprintf(tmp, "%d", tgid);
2599         return vfs_readlink(dentry,buffer,buflen,tmp);
2600 }
2601
2602 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2603 {
2604         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2605         pid_t tgid = task_tgid_nr_ns(current, ns);
2606         char *name = ERR_PTR(-ENOENT);
2607         if (tgid) {
2608                 name = __getname();
2609                 if (!name)
2610                         name = ERR_PTR(-ENOMEM);
2611                 else
2612                         sprintf(name, "%d", tgid);
2613         }
2614         nd_set_link(nd, name);
2615         return NULL;
2616 }
2617
2618 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2619                                 void *cookie)
2620 {
2621         char *s = nd_get_link(nd);
2622         if (!IS_ERR(s))
2623                 __putname(s);
2624 }
2625
2626 static const struct inode_operations proc_self_inode_operations = {
2627         .readlink       = proc_self_readlink,
2628         .follow_link    = proc_self_follow_link,
2629         .put_link       = proc_self_put_link,
2630 };
2631
2632 /*
2633  * proc base
2634  *
2635  * These are the directory entries in the root directory of /proc
2636  * that properly belong to the /proc filesystem, as they describe
2637  * describe something that is process related.
2638  */
2639 static const struct pid_entry proc_base_stuff[] = {
2640         NOD("self", S_IFLNK|S_IRWXUGO,
2641                 &proc_self_inode_operations, NULL, {}),
2642 };
2643
2644 /*
2645  *      Exceptional case: normally we are not allowed to unhash a busy
2646  * directory. In this case, however, we can do it - no aliasing problems
2647  * due to the way we treat inodes.
2648  */
2649 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2650 {
2651         struct inode *inode;
2652         struct task_struct *task;
2653
2654         if (nd->flags & LOOKUP_RCU)
2655                 return -ECHILD;
2656
2657         inode = dentry->d_inode;
2658         task = get_proc_task(inode);
2659         if (task) {
2660                 put_task_struct(task);
2661                 return 1;
2662         }
2663         d_drop(dentry);
2664         return 0;
2665 }
2666
2667 static const struct dentry_operations proc_base_dentry_operations =
2668 {
2669         .d_revalidate   = proc_base_revalidate,
2670         .d_delete       = pid_delete_dentry,
2671 };
2672
2673 static struct dentry *proc_base_instantiate(struct inode *dir,
2674         struct dentry *dentry, struct task_struct *task, const void *ptr)
2675 {
2676         const struct pid_entry *p = ptr;
2677         struct inode *inode;
2678         struct proc_inode *ei;
2679         struct dentry *error;
2680
2681         /* Allocate the inode */
2682         error = ERR_PTR(-ENOMEM);
2683         inode = new_inode(dir->i_sb);
2684         if (!inode)
2685                 goto out;
2686
2687         /* Initialize the inode */
2688         ei = PROC_I(inode);
2689         inode->i_ino = get_next_ino();
2690         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2691
2692         /*
2693          * grab the reference to the task.
2694          */
2695         ei->pid = get_task_pid(task, PIDTYPE_PID);
2696         if (!ei->pid)
2697                 goto out_iput;
2698
2699         inode->i_mode = p->mode;
2700         if (S_ISDIR(inode->i_mode))
2701                 inode->i_nlink = 2;
2702         if (S_ISLNK(inode->i_mode))
2703                 inode->i_size = 64;
2704         if (p->iop)
2705                 inode->i_op = p->iop;
2706         if (p->fop)
2707                 inode->i_fop = p->fop;
2708         ei->op = p->op;
2709         d_set_d_op(dentry, &proc_base_dentry_operations);
2710         d_add(dentry, inode);
2711         error = NULL;
2712 out:
2713         return error;
2714 out_iput:
2715         iput(inode);
2716         goto out;
2717 }
2718
2719 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2720 {
2721         struct dentry *error;
2722         struct task_struct *task = get_proc_task(dir);
2723         const struct pid_entry *p, *last;
2724
2725         error = ERR_PTR(-ENOENT);
2726
2727         if (!task)
2728                 goto out_no_task;
2729
2730         /* Lookup the directory entry */
2731         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2732         for (p = proc_base_stuff; p <= last; p++) {
2733                 if (p->len != dentry->d_name.len)
2734                         continue;
2735                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2736                         break;
2737         }
2738         if (p > last)
2739                 goto out;
2740
2741         error = proc_base_instantiate(dir, dentry, task, p);
2742
2743 out:
2744         put_task_struct(task);
2745 out_no_task:
2746         return error;
2747 }
2748
2749 static int proc_base_fill_cache(struct file *filp, void *dirent,
2750         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2751 {
2752         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2753                                 proc_base_instantiate, task, p);
2754 }
2755
2756 #ifdef CONFIG_TASK_IO_ACCOUNTING
2757 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2758 {
2759         struct task_io_accounting acct = task->ioac;
2760         unsigned long flags;
2761
2762         if (whole && lock_task_sighand(task, &flags)) {
2763                 struct task_struct *t = task;
2764
2765                 task_io_accounting_add(&acct, &task->signal->ioac);
2766                 while_each_thread(task, t)
2767                         task_io_accounting_add(&acct, &t->ioac);
2768
2769                 unlock_task_sighand(task, &flags);
2770         }
2771         return sprintf(buffer,
2772                         "rchar: %llu\n"
2773                         "wchar: %llu\n"
2774                         "syscr: %llu\n"
2775                         "syscw: %llu\n"
2776                         "read_bytes: %llu\n"
2777                         "write_bytes: %llu\n"
2778                         "cancelled_write_bytes: %llu\n",
2779                         (unsigned long long)acct.rchar,
2780                         (unsigned long long)acct.wchar,
2781                         (unsigned long long)acct.syscr,
2782                         (unsigned long long)acct.syscw,
2783                         (unsigned long long)acct.read_bytes,
2784                         (unsigned long long)acct.write_bytes,
2785                         (unsigned long long)acct.cancelled_write_bytes);
2786 }
2787
2788 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2789 {
2790         return do_io_accounting(task, buffer, 0);
2791 }
2792
2793 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2794 {
2795         return do_io_accounting(task, buffer, 1);
2796 }
2797 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2798
2799 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2800                                 struct pid *pid, struct task_struct *task)
2801 {
2802         seq_printf(m, "%08x\n", task->personality);
2803         return 0;
2804 }
2805
2806 /*
2807  * Thread groups
2808  */
2809 static const struct file_operations proc_task_operations;
2810 static const struct inode_operations proc_task_inode_operations;
2811
2812 static const struct pid_entry tgid_base_stuff[] = {
2813         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2814         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2815         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2816 #ifdef CONFIG_NET
2817         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2818 #endif
2819         REG("environ",    S_IRUSR, proc_environ_operations),
2820         INF("auxv",       S_IRUSR, proc_pid_auxv),
2821         ONE("status",     S_IRUGO, proc_pid_status),
2822         ONE("personality", S_IRUSR, proc_pid_personality),
2823         INF("limits",     S_IRUGO, proc_pid_limits),
2824 #ifdef CONFIG_SCHED_DEBUG
2825         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2826 #endif
2827 #ifdef CONFIG_SCHED_AUTOGROUP
2828         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2829 #endif
2830         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2831 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2832         INF("syscall",    S_IRUSR, proc_pid_syscall),
2833 #endif
2834         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2835         ONE("stat",       S_IRUGO, proc_tgid_stat),
2836         ONE("statm",      S_IRUGO, proc_pid_statm),
2837         REG("maps",       S_IRUGO, proc_maps_operations),
2838 #ifdef CONFIG_NUMA
2839         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2840 #endif
2841         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2842         LNK("cwd",        proc_cwd_link),
2843         LNK("root",       proc_root_link),
2844         LNK("exe",        proc_exe_link),
2845         REG("mounts",     S_IRUGO, proc_mounts_operations),
2846         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2847         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2848 #ifdef CONFIG_PROC_PAGE_MONITOR
2849         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2850         REG("smaps",      S_IRUGO, proc_smaps_operations),
2851         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2852 #endif
2853 #ifdef CONFIG_SECURITY
2854         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2855 #endif
2856 #ifdef CONFIG_KALLSYMS
2857         INF("wchan",      S_IRUGO, proc_pid_wchan),
2858 #endif
2859 #ifdef CONFIG_STACKTRACE
2860         ONE("stack",      S_IRUSR, proc_pid_stack),
2861 #endif
2862 #ifdef CONFIG_SCHEDSTATS
2863         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2864 #endif
2865 #ifdef CONFIG_LATENCYTOP
2866         REG("latency",  S_IRUGO, proc_lstats_operations),
2867 #endif
2868 #ifdef CONFIG_PROC_PID_CPUSET
2869         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2870 #endif
2871 #ifdef CONFIG_CGROUPS
2872         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2873 #endif
2874         INF("oom_score",  S_IRUGO, proc_oom_score),
2875         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2876         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2877 #ifdef CONFIG_AUDITSYSCALL
2878         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2879         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2880 #endif
2881 #ifdef CONFIG_FAULT_INJECTION
2882         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2883 #endif
2884 #ifdef CONFIG_ELF_CORE
2885         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2886 #endif
2887 #ifdef CONFIG_TASK_IO_ACCOUNTING
2888         INF("io",       S_IRUGO, proc_tgid_io_accounting),
2889 #endif
2890 };
2891
2892 static int proc_tgid_base_readdir(struct file * filp,
2893                              void * dirent, filldir_t filldir)
2894 {
2895         return proc_pident_readdir(filp,dirent,filldir,
2896                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2897 }
2898
2899 static const struct file_operations proc_tgid_base_operations = {
2900         .read           = generic_read_dir,
2901         .readdir        = proc_tgid_base_readdir,
2902         .llseek         = default_llseek,
2903 };
2904
2905 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2906         return proc_pident_lookup(dir, dentry,
2907                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2908 }
2909
2910 static const struct inode_operations proc_tgid_base_inode_operations = {
2911         .lookup         = proc_tgid_base_lookup,
2912         .getattr        = pid_getattr,
2913         .setattr        = proc_setattr,
2914 };
2915
2916 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2917 {
2918         struct dentry *dentry, *leader, *dir;
2919         char buf[PROC_NUMBUF];
2920         struct qstr name;
2921
2922         name.name = buf;
2923         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2924         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2925         if (dentry) {
2926                 shrink_dcache_parent(dentry);
2927                 d_drop(dentry);
2928                 dput(dentry);
2929         }
2930
2931         name.name = buf;
2932         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2933         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2934         if (!leader)
2935                 goto out;
2936
2937         name.name = "task";
2938         name.len = strlen(name.name);
2939         dir = d_hash_and_lookup(leader, &name);
2940         if (!dir)
2941                 goto out_put_leader;
2942
2943         name.name = buf;
2944         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2945         dentry = d_hash_and_lookup(dir, &name);
2946         if (dentry) {
2947                 shrink_dcache_parent(dentry);
2948                 d_drop(dentry);
2949                 dput(dentry);
2950         }
2951
2952         dput(dir);
2953 out_put_leader:
2954         dput(leader);
2955 out:
2956         return;
2957 }
2958
2959 /**
2960  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2961  * @task: task that should be flushed.
2962  *
2963  * When flushing dentries from proc, one needs to flush them from global
2964  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2965  * in. This call is supposed to do all of this job.
2966  *
2967  * Looks in the dcache for
2968  * /proc/@pid
2969  * /proc/@tgid/task/@pid
2970  * if either directory is present flushes it and all of it'ts children
2971  * from the dcache.
2972  *
2973  * It is safe and reasonable to cache /proc entries for a task until
2974  * that task exits.  After that they just clog up the dcache with
2975  * useless entries, possibly causing useful dcache entries to be
2976  * flushed instead.  This routine is proved to flush those useless
2977  * dcache entries at process exit time.
2978  *
2979  * NOTE: This routine is just an optimization so it does not guarantee
2980  *       that no dcache entries will exist at process exit time it
2981  *       just makes it very unlikely that any will persist.
2982  */
2983
2984 void proc_flush_task(struct task_struct *task)
2985 {
2986         int i;
2987         struct pid *pid, *tgid;
2988         struct upid *upid;
2989
2990         pid = task_pid(task);
2991         tgid = task_tgid(task);
2992
2993         for (i = 0; i <= pid->level; i++) {
2994                 upid = &pid->numbers[i];
2995                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2996                                         tgid->numbers[i].nr);
2997         }
2998
2999         upid = &pid->numbers[pid->level];
3000         if (upid->nr == 1)
3001                 pid_ns_release_proc(upid->ns);
3002 }
3003
3004 static struct dentry *proc_pid_instantiate(struct inode *dir,
3005                                            struct dentry * dentry,
3006                                            struct task_struct *task, const void *ptr)
3007 {
3008         struct dentry *error = ERR_PTR(-ENOENT);
3009         struct inode *inode;
3010
3011         inode = proc_pid_make_inode(dir->i_sb, task);
3012         if (!inode)
3013                 goto out;
3014
3015         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3016         inode->i_op = &proc_tgid_base_inode_operations;
3017         inode->i_fop = &proc_tgid_base_operations;
3018         inode->i_flags|=S_IMMUTABLE;
3019
3020         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
3021                 ARRAY_SIZE(tgid_base_stuff));
3022
3023         d_set_d_op(dentry, &pid_dentry_operations);
3024
3025         d_add(dentry, inode);
3026         /* Close the race of the process dying before we return the dentry */
3027         if (pid_revalidate(dentry, NULL))
3028                 error = NULL;
3029 out:
3030         return error;
3031 }
3032
3033 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3034 {
3035         struct dentry *result;
3036         struct task_struct *task;
3037         unsigned tgid;
3038         struct pid_namespace *ns;
3039
3040         result = proc_base_lookup(dir, dentry);
3041         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3042                 goto out;
3043
3044         tgid = name_to_int(dentry);
3045         if (tgid == ~0U)
3046                 goto out;
3047
3048         ns = dentry->d_sb->s_fs_info;
3049         rcu_read_lock();
3050         task = find_task_by_pid_ns(tgid, ns);
3051         if (task)
3052                 get_task_struct(task);
3053         rcu_read_unlock();
3054         if (!task)
3055                 goto out;
3056
3057         result = proc_pid_instantiate(dir, dentry, task, NULL);
3058         put_task_struct(task);
3059 out:
3060         return result;
3061 }
3062
3063 /*
3064  * Find the first task with tgid >= tgid
3065  *
3066  */
3067 struct tgid_iter {
3068         unsigned int tgid;
3069         struct task_struct *task;
3070 };
3071 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3072 {
3073         struct pid *pid;
3074
3075         if (iter.task)
3076                 put_task_struct(iter.task);
3077         rcu_read_lock();
3078 retry:
3079         iter.task = NULL;
3080         pid = find_ge_pid(iter.tgid, ns);
3081         if (pid) {
3082                 iter.tgid = pid_nr_ns(pid, ns);
3083                 iter.task = pid_task(pid, PIDTYPE_PID);
3084                 /* What we to know is if the pid we have find is the
3085                  * pid of a thread_group_leader.  Testing for task
3086                  * being a thread_group_leader is the obvious thing
3087                  * todo but there is a window when it fails, due to
3088                  * the pid transfer logic in de_thread.
3089                  *
3090                  * So we perform the straight forward test of seeing
3091                  * if the pid we have found is the pid of a thread
3092                  * group leader, and don't worry if the task we have
3093                  * found doesn't happen to be a thread group leader.
3094                  * As we don't care in the case of readdir.
3095                  */
3096                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3097                         iter.tgid += 1;
3098                         goto retry;
3099                 }
3100                 get_task_struct(iter.task);
3101         }
3102         rcu_read_unlock();
3103         return iter;
3104 }
3105
3106 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3107
3108 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3109         struct tgid_iter iter)
3110 {
3111         char name[PROC_NUMBUF];
3112         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3113         return proc_fill_cache(filp, dirent, filldir, name, len,
3114                                 proc_pid_instantiate, iter.task, NULL);
3115 }
3116
3117 /* for the /proc/ directory itself, after non-process stuff has been done */
3118 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3119 {
3120         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3121         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
3122         struct tgid_iter iter;
3123         struct pid_namespace *ns;
3124
3125         if (!reaper)
3126                 goto out_no_task;
3127
3128         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3129                 const struct pid_entry *p = &proc_base_stuff[nr];
3130                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3131                         goto out;
3132         }
3133
3134         ns = filp->f_dentry->d_sb->s_fs_info;
3135         iter.task = NULL;
3136         iter.tgid = filp->f_pos - TGID_OFFSET;
3137         for (iter = next_tgid(ns, iter);
3138              iter.task;
3139              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3140                 filp->f_pos = iter.tgid + TGID_OFFSET;
3141                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
3142                         put_task_struct(iter.task);
3143                         goto out;
3144                 }
3145         }
3146         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3147 out:
3148         put_task_struct(reaper);
3149 out_no_task:
3150         return 0;
3151 }
3152
3153 /*
3154  * Tasks
3155  */
3156 static const struct pid_entry tid_base_stuff[] = {
3157         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3158         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3159         REG("environ",   S_IRUSR, proc_environ_operations),
3160         INF("auxv",      S_IRUSR, proc_pid_auxv),
3161         ONE("status",    S_IRUGO, proc_pid_status),
3162         ONE("personality", S_IRUSR, proc_pid_personality),
3163         INF("limits",    S_IRUGO, proc_pid_limits),
3164 #ifdef CONFIG_SCHED_DEBUG
3165         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3166 #endif
3167         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3168 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3169         INF("syscall",   S_IRUSR, proc_pid_syscall),
3170 #endif
3171         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3172         ONE("stat",      S_IRUGO, proc_tid_stat),
3173         ONE("statm",     S_IRUGO, proc_pid_statm),
3174         REG("maps",      S_IRUGO, proc_maps_operations),
3175 #ifdef CONFIG_NUMA
3176         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
3177 #endif
3178         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3179         LNK("cwd",       proc_cwd_link),
3180         LNK("root",      proc_root_link),
3181         LNK("exe",       proc_exe_link),
3182         REG("mounts",    S_IRUGO, proc_mounts_operations),
3183         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3184 #ifdef CONFIG_PROC_PAGE_MONITOR
3185         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3186         REG("smaps",     S_IRUGO, proc_smaps_operations),
3187         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3188 #endif
3189 #ifdef CONFIG_SECURITY
3190         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3191 #endif
3192 #ifdef CONFIG_KALLSYMS
3193         INF("wchan",     S_IRUGO, proc_pid_wchan),
3194 #endif
3195 #ifdef CONFIG_STACKTRACE
3196         ONE("stack",      S_IRUSR, proc_pid_stack),
3197 #endif
3198 #ifdef CONFIG_SCHEDSTATS
3199         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3200 #endif
3201 #ifdef CONFIG_LATENCYTOP
3202         REG("latency",  S_IRUGO, proc_lstats_operations),
3203 #endif
3204 #ifdef CONFIG_PROC_PID_CPUSET
3205         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3206 #endif
3207 #ifdef CONFIG_CGROUPS
3208         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3209 #endif
3210         INF("oom_score", S_IRUGO, proc_oom_score),
3211         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3212         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3213 #ifdef CONFIG_AUDITSYSCALL
3214         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3215         REG("sessionid",  S_IRUSR, proc_sessionid_operations),
3216 #endif
3217 #ifdef CONFIG_FAULT_INJECTION
3218         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3219 #endif
3220 #ifdef CONFIG_TASK_IO_ACCOUNTING
3221         INF("io",       S_IRUGO, proc_tid_io_accounting),
3222 #endif
3223 };
3224
3225 static int proc_tid_base_readdir(struct file * filp,
3226                              void * dirent, filldir_t filldir)
3227 {
3228         return proc_pident_readdir(filp,dirent,filldir,
3229                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3230 }
3231
3232 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3233         return proc_pident_lookup(dir, dentry,
3234                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3235 }
3236
3237 static const struct file_operations proc_tid_base_operations = {
3238         .read           = generic_read_dir,
3239         .readdir        = proc_tid_base_readdir,
3240         .llseek         = default_llseek,
3241 };
3242
3243 static const struct inode_operations proc_tid_base_inode_operations = {
3244         .lookup         = proc_tid_base_lookup,
3245         .getattr        = pid_getattr,
3246         .setattr        = proc_setattr,
3247 };
3248
3249 static struct dentry *proc_task_instantiate(struct inode *dir,
3250         struct dentry *dentry, struct task_struct *task, const void *ptr)
3251 {
3252         struct dentry *error = ERR_PTR(-ENOENT);
3253         struct inode *inode;
3254         inode = proc_pid_make_inode(dir->i_sb, task);
3255
3256         if (!inode)
3257                 goto out;
3258         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3259         inode->i_op = &proc_tid_base_inode_operations;
3260         inode->i_fop = &proc_tid_base_operations;
3261         inode->i_flags|=S_IMMUTABLE;
3262
3263         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
3264                 ARRAY_SIZE(tid_base_stuff));
3265
3266         d_set_d_op(dentry, &pid_dentry_operations);
3267
3268         d_add(dentry, inode);
3269         /* Close the race of the process dying before we return the dentry */
3270         if (pid_revalidate(dentry, NULL))
3271                 error = NULL;
3272 out:
3273         return error;
3274 }
3275
3276 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3277 {
3278         struct dentry *result = ERR_PTR(-ENOENT);
3279         struct task_struct *task;
3280         struct task_struct *leader = get_proc_task(dir);
3281         unsigned tid;
3282         struct pid_namespace *ns;
3283
3284         if (!leader)
3285                 goto out_no_task;
3286
3287         tid = name_to_int(dentry);
3288         if (tid == ~0U)
3289                 goto out;
3290
3291         ns = dentry->d_sb->s_fs_info;
3292         rcu_read_lock();
3293         task = find_task_by_pid_ns(tid, ns);
3294         if (task)
3295                 get_task_struct(task);
3296         rcu_read_unlock();
3297         if (!task)
3298                 goto out;
3299         if (!same_thread_group(leader, task))
3300                 goto out_drop_task;
3301
3302         result = proc_task_instantiate(dir, dentry, task, NULL);
3303 out_drop_task:
3304         put_task_struct(task);
3305 out:
3306         put_task_struct(leader);
3307 out_no_task:
3308         return result;
3309 }
3310
3311 /*
3312  * Find the first tid of a thread group to return to user space.
3313  *
3314  * Usually this is just the thread group leader, but if the users
3315  * buffer was too small or there was a seek into the middle of the
3316  * directory we have more work todo.
3317  *
3318  * In the case of a short read we start with find_task_by_pid.
3319  *
3320  * In the case of a seek we start with the leader and walk nr
3321  * threads past it.
3322  */
3323 static struct task_struct *first_tid(struct task_struct *leader,
3324                 int tid, int nr, struct pid_namespace *ns)
3325 {
3326         struct task_struct *pos;
3327
3328         rcu_read_lock();
3329         /* Attempt to start with the pid of a thread */
3330         if (tid && (nr > 0)) {
3331                 pos = find_task_by_pid_ns(tid, ns);
3332                 if (pos && (pos->group_leader == leader))
3333                         goto found;
3334         }
3335
3336         /* If nr exceeds the number of threads there is nothing todo */
3337         pos = NULL;
3338         if (nr && nr >= get_nr_threads(leader))
3339                 goto out;
3340
3341         /* If we haven't found our starting place yet start
3342          * with the leader and walk nr threads forward.
3343          */
3344         for (pos = leader; nr > 0; --nr) {
3345                 pos = next_thread(pos);
3346                 if (pos == leader) {
3347                         pos = NULL;
3348                         goto out;
3349                 }
3350         }
3351 found:
3352         get_task_struct(pos);
3353 out:
3354         rcu_read_unlock();
3355         return pos;
3356 }
3357
3358 /*
3359  * Find the next thread in the thread list.
3360  * Return NULL if there is an error or no next thread.
3361  *
3362  * The reference to the input task_struct is released.
3363  */
3364 static struct task_struct *next_tid(struct task_struct *start)
3365 {
3366         struct task_struct *pos = NULL;
3367         rcu_read_lock();
3368         if (pid_alive(start)) {
3369                 pos = next_thread(start);
3370                 if (thread_group_leader(pos))
3371                         pos = NULL;
3372                 else
3373                         get_task_struct(pos);
3374         }
3375         rcu_read_unlock();
3376         put_task_struct(start);
3377         return pos;
3378 }
3379
3380 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3381         struct task_struct *task, int tid)
3382 {
3383         char name[PROC_NUMBUF];
3384         int len = snprintf(name, sizeof(name), "%d", tid);
3385         return proc_fill_cache(filp, dirent, filldir, name, len,
3386                                 proc_task_instantiate, task, NULL);
3387 }
3388
3389 /* for the /proc/TGID/task/ directories */
3390 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3391 {
3392         struct dentry *dentry = filp->f_path.dentry;
3393         struct inode *inode = dentry->d_inode;
3394         struct task_struct *leader = NULL;
3395         struct task_struct *task;
3396         int retval = -ENOENT;
3397         ino_t ino;
3398         int tid;
3399         struct pid_namespace *ns;
3400
3401         task = get_proc_task(inode);
3402         if (!task)
3403                 goto out_no_task;
3404         rcu_read_lock();
3405         if (pid_alive(task)) {
3406                 leader = task->group_leader;
3407                 get_task_struct(leader);
3408         }
3409         rcu_read_unlock();
3410         put_task_struct(task);
3411         if (!leader)
3412                 goto out_no_task;
3413         retval = 0;
3414
3415         switch ((unsigned long)filp->f_pos) {
3416         case 0:
3417                 ino = inode->i_ino;
3418                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3419                         goto out;
3420                 filp->f_pos++;
3421                 /* fall through */
3422         case 1:
3423                 ino = parent_ino(dentry);
3424                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3425                         goto out;
3426                 filp->f_pos++;
3427                 /* fall through */
3428         }
3429
3430         /* f_version caches the tgid value that the last readdir call couldn't
3431          * return. lseek aka telldir automagically resets f_version to 0.
3432          */
3433         ns = filp->f_dentry->d_sb->s_fs_info;
3434         tid = (int)filp->f_version;
3435         filp->f_version = 0;
3436         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3437              task;
3438              task = next_tid(task), filp->f_pos++) {
3439                 tid = task_pid_nr_ns(task, ns);
3440                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3441                         /* returning this tgid failed, save it as the first
3442                          * pid for the next readir call */
3443                         filp->f_version = (u64)tid;
3444                         put_task_struct(task);
3445                         break;
3446                 }
3447         }
3448 out:
3449         put_task_struct(leader);
3450 out_no_task:
3451         return retval;
3452 }
3453
3454 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3455 {
3456         struct inode *inode = dentry->d_inode;
3457         struct task_struct *p = get_proc_task(inode);
3458         generic_fillattr(inode, stat);
3459
3460         if (p) {
3461                 stat->nlink += get_nr_threads(p);
3462                 put_task_struct(p);
3463         }
3464
3465         return 0;
3466 }
3467
3468 static const struct inode_operations proc_task_inode_operations = {
3469         .lookup         = proc_task_lookup,
3470         .getattr        = proc_task_getattr,
3471         .setattr        = proc_setattr,
3472 };
3473
3474 static const struct file_operations proc_task_operations = {
3475         .read           = generic_read_dir,
3476         .readdir        = proc_task_readdir,
3477         .llseek         = default_llseek,
3478 };