fs/proc/base.c

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