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