fs/proc/base.c

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