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