SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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/rcupdate.h>
  67 #include <linux/kallsyms.h>
  68 #include <linux/resource.h>
  69 #include <linux/module.h>
  70 #include <linux/mount.h>
  71 #include <linux/security.h>
  72 #include <linux/ptrace.h>
  73 #include <linux/tracehook.h>
  74 #include <linux/cgroup.h>
  75 #include <linux/cpuset.h>
  76 #include <linux/audit.h>
  77 #include <linux/poll.h>
  78 #include <linux/nsproxy.h>
  79 #include <linux/oom.h>
  80 #include <linux/elf.h>
  81 #include <linux/pid_namespace.h>
  82 #include "internal.h"
  83
  84 /* NOTE:
  85  *      Implementing inode permission operations in /proc is almost
  86  *      certainly an error.  Permission checks need to happen during
  87  *      each system call not at open time.  The reason is that most of
  88  *      what we wish to check for permissions in /proc varies at runtime.
  89  *
  90  *      The classic example of a problem is opening file descriptors
  91  *      in /proc for a task before it execs a suid executable.
  92  */
  93
  94 struct pid_entry {
  95         char *name;
  96         int len;
  97         mode_t mode;
  98         const struct inode_operations *iop;
  99         const struct file_operations *fop;
 100         union proc_op op;
 101 };
 102
 103 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 104         .name = (NAME),                                 \
 105         .len  = sizeof(NAME) - 1,                       \
 106         .mode = MODE,                                   \
 107         .iop  = IOP,                                    \
 108         .fop  = FOP,                                    \
 109         .op   = OP,                                     \
 110 }
 111
 112 #define DIR(NAME, MODE, OTYPE)                                                  \
 113         NOD(NAME, (S_IFDIR|(MODE)),                                             \
 114                 &proc_##OTYPE##_inode_operations, &proc_##OTYPE##_operations,   \
 115                 {} )
 116 #define LNK(NAME, OTYPE)                                        \
 117         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 118                 &proc_pid_link_inode_operations, NULL,          \
 119                 { .proc_get_link = &proc_##OTYPE##_link } )
 120 #define REG(NAME, MODE, OTYPE)                          \
 121         NOD(NAME, (S_IFREG|(MODE)), NULL,               \
 122                 &proc_##OTYPE##_operations, {})
 123 #define INF(NAME, MODE, OTYPE)                          \
 124         NOD(NAME, (S_IFREG|(MODE)),                     \
 125                 NULL, &proc_info_file_operations,       \
 126                 { .proc_read = &proc_##OTYPE } )
 127 #define ONE(NAME, MODE, OTYPE)                          \
 128         NOD(NAME, (S_IFREG|(MODE)),                     \
 129                 NULL, &proc_single_file_operations,     \
 130                 { .proc_show = &proc_##OTYPE } )
 131
 132 /*
 133  * Count the number of hardlinks for the pid_entry table, excluding the .
 134  * and .. links.
 135  */
 136 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 137         unsigned int n)
 138 {
 139         unsigned int i;
 140         unsigned int count;
 141
 142         count = 0;
 143         for (i = 0; i < n; ++i) {
 144                 if (S_ISDIR(entries[i].mode))
 145                         ++count;
 146         }
 147
 148         return count;
 149 }
 150
 151 static struct fs_struct *get_fs_struct(struct task_struct *task)
 152 {
 153         struct fs_struct *fs;
 154         task_lock(task);
 155         fs = task->fs;
 156         if(fs)
 157                 atomic_inc(&fs->count);
 158         task_unlock(task);
 159         return fs;
 160 }
 161
 162 static int get_nr_threads(struct task_struct *tsk)
 163 {
 164         unsigned long flags;
 165         int count = 0;
 166
 167         if (lock_task_sighand(tsk, &flags)) {
 168                 count = atomic_read(&tsk->signal->count);
 169                 unlock_task_sighand(tsk, &flags);
 170         }
 171         return count;
 172 }
 173
 174 static int proc_cwd_link(struct inode *inode, struct path *path)
 175 {
 176         struct task_struct *task = get_proc_task(inode);
 177         struct fs_struct *fs = NULL;
 178         int result = -ENOENT;
 179
 180         if (task) {
 181                 fs = get_fs_struct(task);
 182                 put_task_struct(task);
 183         }
 184         if (fs) {
 185                 read_lock(&fs->lock);
 186                 *path = fs->pwd;
 187                 path_get(&fs->pwd);
 188                 read_unlock(&fs->lock);
 189                 result = 0;
 190                 put_fs_struct(fs);
 191         }
 192         return result;
 193 }
 194
 195 static int proc_root_link(struct inode *inode, struct path *path)
 196 {
 197         struct task_struct *task = get_proc_task(inode);
 198         struct fs_struct *fs = NULL;
 199         int result = -ENOENT;
 200
 201         if (task) {
 202                 fs = get_fs_struct(task);
 203                 put_task_struct(task);
 204         }
 205         if (fs) {
 206                 read_lock(&fs->lock);
 207                 *path = fs->root;
 208                 path_get(&fs->root);
 209                 read_unlock(&fs->lock);
 210                 result = 0;
 211                 put_fs_struct(fs);
 212         }
 213         return result;
 214 }
 215
 216 /*
 217  * Return zero if current may access user memory in @task, -error if not.
 218  */
 219 static int check_mem_permission(struct task_struct *task)
 220 {
 221         /*
 222          * A task can always look at itself, in case it chooses
 223          * to use system calls instead of load instructions.
 224          */
 225         if (task == current)
 226                 return 0;
 227
 228         /*
 229          * If current is actively ptrace'ing, and would also be
 230          * permitted to freshly attach with ptrace now, permit it.
 231          */
 232         if (task_is_stopped_or_traced(task)) {
 233                 int match;
 234                 rcu_read_lock();
 235                 match = (tracehook_tracer_task(task) == current);
 236                 rcu_read_unlock();
 237                 if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH))
 238                         return 0;
 239         }
 240
 241         /*
 242          * Noone else is allowed.
 243          */
 244         return -EPERM;
 245 }
 246
 247 struct mm_struct *mm_for_maps(struct task_struct *task)
 248 {
 249         struct mm_struct *mm = get_task_mm(task);
 250         if (!mm)
 251                 return NULL;
 252         down_read(&mm->mmap_sem);
 253         task_lock(task);
 254         if (task->mm != mm)
 255                 goto out;
 256         if (task->mm != current->mm &&
 257             __ptrace_may_access(task, PTRACE_MODE_READ) < 0)
 258                 goto out;
 259         task_unlock(task);
 260         return mm;
 261 out:
 262         task_unlock(task);
 263         up_read(&mm->mmap_sem);
 264         mmput(mm);
 265         return NULL;
 266 }
 267
 268 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 269 {
 270         int res = 0;
 271         unsigned int len;
 272         struct mm_struct *mm = get_task_mm(task);
 273         if (!mm)
 274                 goto out;
 275         if (!mm->arg_end)
 276                 goto out_mm;    /* Shh! No looking before we're done */
 277
 278         len = mm->arg_end - mm->arg_start;
 279
 280         if (len > PAGE_SIZE)
 281                 len = PAGE_SIZE;
 282
 283         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 284
 285         // If the nul at the end of args has been overwritten, then
 286         // assume application is using setproctitle(3).
 287         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 288                 len = strnlen(buffer, res);
 289                 if (len < res) {
 290                     res = len;
 291                 } else {
 292                         len = mm->env_end - mm->env_start;
 293                         if (len > PAGE_SIZE - res)
 294                                 len = PAGE_SIZE - res;
 295                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 296                         res = strnlen(buffer, res);
 297                 }
 298         }
 299 out_mm:
 300         mmput(mm);
 301 out:
 302         return res;
 303 }
 304
 305 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 306 {
 307         int res = 0;
 308         struct mm_struct *mm = get_task_mm(task);
 309         if (mm) {
 310                 unsigned int nwords = 0;
 311                 do
 312                         nwords += 2;
 313                 while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 314                 res = nwords * sizeof(mm->saved_auxv[0]);
 315                 if (res > PAGE_SIZE)
 316                         res = PAGE_SIZE;
 317                 memcpy(buffer, mm->saved_auxv, res);
 318                 mmput(mm);
 319         }
 320         return res;
 321 }
 322
 323
 324 #ifdef CONFIG_KALLSYMS
 325 /*
 326  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 327  * Returns the resolved symbol.  If that fails, simply return the address.
 328  */
 329 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 330 {
 331         unsigned long wchan;
 332         char symname[KSYM_NAME_LEN];
 333
 334         wchan = get_wchan(task);
 335
 336         if (lookup_symbol_name(wchan, symname) < 0)
 337                 return sprintf(buffer, "%lu", wchan);
 338         else
 339                 return sprintf(buffer, "%s", symname);
 340 }
 341 #endif /* CONFIG_KALLSYMS */
 342
 343 #ifdef CONFIG_SCHEDSTATS
 344 /*
 345  * Provides /proc/PID/schedstat
 346  */
 347 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 348 {
 349         return sprintf(buffer, "%llu %llu %lu\n",
 350                         (unsigned long long)task->se.sum_exec_runtime,
 351                         (unsigned long long)task->sched_info.run_delay,
 352                         task->sched_info.pcount);
 353 }
 354 #endif
 355
 356 #ifdef CONFIG_LATENCYTOP
 357 static int lstats_show_proc(struct seq_file *m, void *v)
 358 {
 359         int i;
 360         struct inode *inode = m->private;
 361         struct task_struct *task = get_proc_task(inode);
 362
 363         if (!task)
 364                 return -ESRCH;
 365         seq_puts(m, "Latency Top version : v0.1\n");
 366         for (i = 0; i < 32; i++) {
 367                 if (task->latency_record[i].backtrace[0]) {
 368                         int q;
 369                         seq_printf(m, "%i %li %li ",
 370                                 task->latency_record[i].count,
 371                                 task->latency_record[i].time,
 372                                 task->latency_record[i].max);
 373                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 374                                 char sym[KSYM_SYMBOL_LEN];
 375                                 char *c;
 376                                 if (!task->latency_record[i].backtrace[q])
 377                                         break;
 378                                 if (task->latency_record[i].backtrace[q] == ULONG_MAX)
 379                                         break;
 380                                 sprint_symbol(sym, task->latency_record[i].backtrace[q]);
 381                                 c = strchr(sym, '+');
 382                                 if (c)
 383                                         *c = 0;
 384                                 seq_printf(m, "%s ", sym);
 385                         }
 386                         seq_printf(m, "\n");
 387                 }
 388
 389         }
 390         put_task_struct(task);
 391         return 0;
 392 }
 393
 394 static int lstats_open(struct inode *inode, struct file *file)
 395 {
 396         return single_open(file, lstats_show_proc, inode);
 397 }
 398
 399 static ssize_t lstats_write(struct file *file, const char __user *buf,
 400                             size_t count, loff_t *offs)
 401 {
 402         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 403
 404         if (!task)
 405                 return -ESRCH;
 406         clear_all_latency_tracing(task);
 407         put_task_struct(task);
 408
 409         return count;
 410 }
 411
 412 static const struct file_operations proc_lstats_operations = {
 413         .open           = lstats_open,
 414         .read           = seq_read,
 415         .write          = lstats_write,
 416         .llseek         = seq_lseek,
 417         .release        = single_release,
 418 };
 419
 420 #endif
 421
 422 /* The badness from the OOM killer */
 423 unsigned long badness(struct task_struct *p, unsigned long uptime);
 424 static int proc_oom_score(struct task_struct *task, char *buffer)
 425 {
 426         unsigned long points;
 427         struct timespec uptime;
 428
 429         do_posix_clock_monotonic_gettime(&uptime);
 430         read_lock(&tasklist_lock);
 431         points = badness(task, uptime.tv_sec);
 432         read_unlock(&tasklist_lock);
 433         return sprintf(buffer, "%lu\n", points);
 434 }
 435
 436 struct limit_names {
 437         char *name;
 438         char *unit;
 439 };
 440
 441 static const struct limit_names lnames[RLIM_NLIMITS] = {
 442         [RLIMIT_CPU] = {"Max cpu time", "ms"},
 443         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 444         [RLIMIT_DATA] = {"Max data size", "bytes"},
 445         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 446         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 447         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 448         [RLIMIT_NPROC] = {"Max processes", "processes"},
 449         [RLIMIT_NOFILE] = {"Max open files", "files"},
 450         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 451         [RLIMIT_AS] = {"Max address space", "bytes"},
 452         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 453         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 454         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 455         [RLIMIT_NICE] = {"Max nice priority", NULL},
 456         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 457         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 458 };
 459
 460 /* Display limits for a process */
 461 static int proc_pid_limits(struct task_struct *task, char *buffer)
 462 {
 463         unsigned int i;
 464         int count = 0;
 465         unsigned long flags;
 466         char *bufptr = buffer;
 467
 468         struct rlimit rlim[RLIM_NLIMITS];
 469
 470         if (!lock_task_sighand(task, &flags))
 471                 return 0;
 472         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 473         unlock_task_sighand(task, &flags);
 474
 475         /*
 476          * print the file header
 477          */
 478         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 479                         "Limit", "Soft Limit", "Hard Limit", "Units");
 480
 481         for (i = 0; i < RLIM_NLIMITS; i++) {
 482                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 483                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 484                                          lnames[i].name, "unlimited");
 485                 else
 486                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 487                                          lnames[i].name, rlim[i].rlim_cur);
 488
 489                 if (rlim[i].rlim_max == RLIM_INFINITY)
 490                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 491                 else
 492                         count += sprintf(&bufptr[count], "%-20lu ",
 493                                          rlim[i].rlim_max);
 494
 495                 if (lnames[i].unit)
 496                         count += sprintf(&bufptr[count], "%-10s\n",
 497                                          lnames[i].unit);
 498                 else
 499                         count += sprintf(&bufptr[count], "\n");
 500         }
 501
 502         return count;
 503 }
 504
 505 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 506 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 507 {
 508         long nr;
 509         unsigned long args[6], sp, pc;
 510
 511         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 512                 return sprintf(buffer, "running\n");
 513
 514         if (nr < 0)
 515                 return sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 516
 517         return sprintf(buffer,
 518                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 519                        nr,
 520                        args[0], args[1], args[2], args[3], args[4], args[5],
 521                        sp, pc);
 522 }
 523 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 524
 525 /************************************************************************/
 526 /*                       Here the fs part begins                        */
 527 /************************************************************************/
 528
 529 /* permission checks */
 530 static int proc_fd_access_allowed(struct inode *inode)
 531 {
 532         struct task_struct *task;
 533         int allowed = 0;
 534         /* Allow access to a task's file descriptors if it is us or we
 535          * may use ptrace attach to the process and find out that
 536          * information.
 537          */
 538         task = get_proc_task(inode);
 539         if (task) {
 540                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 541                 put_task_struct(task);
 542         }
 543         return allowed;
 544 }
 545
 546 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
 547 {
 548         int error;
 549         struct inode *inode = dentry->d_inode;
 550
 551         if (attr->ia_valid & ATTR_MODE)
 552                 return -EPERM;
 553
 554         error = inode_change_ok(inode, attr);
 555         if (!error)
 556                 error = inode_setattr(inode, attr);
 557         return error;
 558 }
 559
 560 static const struct inode_operations proc_def_inode_operations = {
 561         .setattr        = proc_setattr,
 562 };
 563
 564 static int mounts_open_common(struct inode *inode, struct file *file,
 565                               const struct seq_operations *op)
 566 {
 567         struct task_struct *task = get_proc_task(inode);
 568         struct nsproxy *nsp;
 569         struct mnt_namespace *ns = NULL;
 570         struct fs_struct *fs = NULL;
 571         struct path root;
 572         struct proc_mounts *p;
 573         int ret = -EINVAL;
 574
 575         if (task) {
 576                 rcu_read_lock();
 577                 nsp = task_nsproxy(task);
 578                 if (nsp) {
 579                         ns = nsp->mnt_ns;
 580                         if (ns)
 581                                 get_mnt_ns(ns);
 582                 }
 583                 rcu_read_unlock();
 584                 if (ns)
 585                         fs = get_fs_struct(task);
 586                 put_task_struct(task);
 587         }
 588
 589         if (!ns)
 590                 goto err;
 591         if (!fs)
 592                 goto err_put_ns;
 593
 594         read_lock(&fs->lock);
 595         root = fs->root;
 596         path_get(&root);
 597         read_unlock(&fs->lock);
 598         put_fs_struct(fs);
 599
 600         ret = -ENOMEM;
 601         p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 602         if (!p)
 603                 goto err_put_path;
 604
 605         file->private_data = &p->m;
 606         ret = seq_open(file, op);
 607         if (ret)
 608                 goto err_free;
 609
 610         p->m.private = p;
 611         p->ns = ns;
 612         p->root = root;
 613         p->event = ns->event;
 614
 615         return 0;
 616
 617  err_free:
 618         kfree(p);
 619  err_put_path:
 620         path_put(&root);
 621  err_put_ns:
 622         put_mnt_ns(ns);
 623  err:
 624         return ret;
 625 }
 626
 627 static int mounts_release(struct inode *inode, struct file *file)
 628 {
 629         struct proc_mounts *p = file->private_data;
 630         path_put(&p->root);
 631         put_mnt_ns(p->ns);
 632         return seq_release(inode, file);
 633 }
 634
 635 static unsigned mounts_poll(struct file *file, poll_table *wait)
 636 {
 637         struct proc_mounts *p = file->private_data;
 638         struct mnt_namespace *ns = p->ns;
 639         unsigned res = 0;
 640
 641         poll_wait(file, &ns->poll, wait);
 642
 643         spin_lock(&vfsmount_lock);
 644         if (p->event != ns->event) {
 645                 p->event = ns->event;
 646                 res = POLLERR;
 647         }
 648         spin_unlock(&vfsmount_lock);
 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 };
 726
 727 static int proc_single_show(struct seq_file *m, void *v)
 728 {
 729         struct inode *inode = m->private;
 730         struct pid_namespace *ns;
 731         struct pid *pid;
 732         struct task_struct *task;
 733         int ret;
 734
 735         ns = inode->i_sb->s_fs_info;
 736         pid = proc_pid(inode);
 737         task = get_pid_task(pid, PIDTYPE_PID);
 738         if (!task)
 739                 return -ESRCH;
 740
 741         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 742
 743         put_task_struct(task);
 744         return ret;
 745 }
 746
 747 static int proc_single_open(struct inode *inode, struct file *filp)
 748 {
 749         int ret;
 750         ret = single_open(filp, proc_single_show, NULL);
 751         if (!ret) {
 752                 struct seq_file *m = filp->private_data;
 753
 754                 m->private = inode;
 755         }
 756         return ret;
 757 }
 758
 759 static const struct file_operations proc_single_file_operations = {
 760         .open           = proc_single_open,
 761         .read           = seq_read,
 762         .llseek         = seq_lseek,
 763         .release        = single_release,
 764 };
 765
 766 static int mem_open(struct inode* inode, struct file* file)
 767 {
 768         file->private_data = (void*)((long)current->self_exec_id);
 769         return 0;
 770 }
 771
 772 static ssize_t mem_read(struct file * file, char __user * buf,
 773                         size_t count, loff_t *ppos)
 774 {
 775         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 776         char *page;
 777         unsigned long src = *ppos;
 778         int ret = -ESRCH;
 779         struct mm_struct *mm;
 780
 781         if (!task)
 782                 goto out_no_task;
 783
 784         if (check_mem_permission(task))
 785                 goto out;
 786
 787         ret = -ENOMEM;
 788         page = (char *)__get_free_page(GFP_TEMPORARY);
 789         if (!page)
 790                 goto out;
 791
 792         ret = 0;
 793
 794         mm = get_task_mm(task);
 795         if (!mm)
 796                 goto out_free;
 797
 798         ret = -EIO;
 799
 800         if (file->private_data != (void*)((long)current->self_exec_id))
 801                 goto out_put;
 802
 803         ret = 0;
 804
 805         while (count > 0) {
 806                 int this_len, retval;
 807
 808                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 809                 retval = access_process_vm(task, src, page, this_len, 0);
 810                 if (!retval || check_mem_permission(task)) {
 811                         if (!ret)
 812                                 ret = -EIO;
 813                         break;
 814                 }
 815
 816                 if (copy_to_user(buf, page, retval)) {
 817                         ret = -EFAULT;
 818                         break;
 819                 }
 820
 821                 ret += retval;
 822                 src += retval;
 823                 buf += retval;
 824                 count -= retval;
 825         }
 826         *ppos = src;
 827
 828 out_put:
 829         mmput(mm);
 830 out_free:
 831         free_page((unsigned long) page);
 832 out:
 833         put_task_struct(task);
 834 out_no_task:
 835         return ret;
 836 }
 837
 838 #define mem_write NULL
 839
 840 #ifndef mem_write
 841 /* This is a security hazard */
 842 static ssize_t mem_write(struct file * file, const char __user *buf,
 843                          size_t count, loff_t *ppos)
 844 {
 845         int copied;
 846         char *page;
 847         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 848         unsigned long dst = *ppos;
 849
 850         copied = -ESRCH;
 851         if (!task)
 852                 goto out_no_task;
 853
 854         if (check_mem_permission(task))
 855                 goto out;
 856
 857         copied = -ENOMEM;
 858         page = (char *)__get_free_page(GFP_TEMPORARY);
 859         if (!page)
 860                 goto out;
 861
 862         copied = 0;
 863         while (count > 0) {
 864                 int this_len, retval;
 865
 866                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 867                 if (copy_from_user(page, buf, this_len)) {
 868                         copied = -EFAULT;
 869                         break;
 870                 }
 871                 retval = access_process_vm(task, dst, page, this_len, 1);
 872                 if (!retval) {
 873                         if (!copied)
 874                                 copied = -EIO;
 875                         break;
 876                 }
 877                 copied += retval;
 878                 buf += retval;
 879                 dst += retval;
 880                 count -= retval;
 881         }
 882         *ppos = dst;
 883         free_page((unsigned long) page);
 884 out:
 885         put_task_struct(task);
 886 out_no_task:
 887         return copied;
 888 }
 889 #endif
 890
 891 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 892 {
 893         switch (orig) {
 894         case 0:
 895                 file->f_pos = offset;
 896                 break;
 897         case 1:
 898                 file->f_pos += offset;
 899                 break;
 900         default:
 901                 return -EINVAL;
 902         }
 903         force_successful_syscall_return();
 904         return file->f_pos;
 905 }
 906
 907 static const struct file_operations proc_mem_operations = {
 908         .llseek         = mem_lseek,
 909         .read           = mem_read,
 910         .write          = mem_write,
 911         .open           = mem_open,
 912 };
 913
 914 static ssize_t environ_read(struct file *file, char __user *buf,
 915                         size_t count, loff_t *ppos)
 916 {
 917         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 918         char *page;
 919         unsigned long src = *ppos;
 920         int ret = -ESRCH;
 921         struct mm_struct *mm;
 922
 923         if (!task)
 924                 goto out_no_task;
 925
 926         if (!ptrace_may_access(task, PTRACE_MODE_READ))
 927                 goto out;
 928
 929         ret = -ENOMEM;
 930         page = (char *)__get_free_page(GFP_TEMPORARY);
 931         if (!page)
 932                 goto out;
 933
 934         ret = 0;
 935
 936         mm = get_task_mm(task);
 937         if (!mm)
 938                 goto out_free;
 939
 940         while (count > 0) {
 941                 int this_len, retval, max_len;
 942
 943                 this_len = mm->env_end - (mm->env_start + src);
 944
 945                 if (this_len <= 0)
 946                         break;
 947
 948                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 949                 this_len = (this_len > max_len) ? max_len : this_len;
 950
 951                 retval = access_process_vm(task, (mm->env_start + src),
 952                         page, this_len, 0);
 953
 954                 if (retval <= 0) {
 955                         ret = retval;
 956                         break;
 957                 }
 958
 959                 if (copy_to_user(buf, page, retval)) {
 960                         ret = -EFAULT;
 961                         break;
 962                 }
 963
 964                 ret += retval;
 965                 src += retval;
 966                 buf += retval;
 967                 count -= retval;
 968         }
 969         *ppos = src;
 970
 971         mmput(mm);
 972 out_free:
 973         free_page((unsigned long) page);
 974 out:
 975         put_task_struct(task);
 976 out_no_task:
 977         return ret;
 978 }
 979
 980 static const struct file_operations proc_environ_operations = {
 981         .read           = environ_read,
 982 };
 983
 984 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 985                                 size_t count, loff_t *ppos)
 986 {
 987         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 988         char buffer[PROC_NUMBUF];
 989         size_t len;
 990         int oom_adjust;
 991
 992         if (!task)
 993                 return -ESRCH;
 994         oom_adjust = task->oomkilladj;
 995         put_task_struct(task);
 996
 997         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
 998
 999         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1000 }
1001
1002 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1003                                 size_t count, loff_t *ppos)
1004 {
1005         struct task_struct *task;
1006         char buffer[PROC_NUMBUF], *end;
1007         int oom_adjust;
1008
1009         memset(buffer, 0, sizeof(buffer));
1010         if (count > sizeof(buffer) - 1)
1011                 count = sizeof(buffer) - 1;
1012         if (copy_from_user(buffer, buf, count))
1013                 return -EFAULT;
1014         oom_adjust = simple_strtol(buffer, &end, 0);
1015         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1016              oom_adjust != OOM_DISABLE)
1017                 return -EINVAL;
1018         if (*end == '\n')
1019                 end++;
1020         task = get_proc_task(file->f_path.dentry->d_inode);
1021         if (!task)
1022                 return -ESRCH;
1023         if (oom_adjust < task->oomkilladj && !capable(CAP_SYS_RESOURCE)) {
1024                 put_task_struct(task);
1025                 return -EACCES;
1026         }
1027         task->oomkilladj = oom_adjust;
1028         put_task_struct(task);
1029         if (end - buffer == 0)
1030                 return -EIO;
1031         return end - buffer;
1032 }
1033
1034 static const struct file_operations proc_oom_adjust_operations = {
1035         .read           = oom_adjust_read,
1036         .write          = oom_adjust_write,
1037 };
1038
1039 #ifdef CONFIG_AUDITSYSCALL
1040 #define TMPBUFLEN 21
1041 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1042                                   size_t count, loff_t *ppos)
1043 {
1044         struct inode * inode = file->f_path.dentry->d_inode;
1045         struct task_struct *task = get_proc_task(inode);
1046         ssize_t length;
1047         char tmpbuf[TMPBUFLEN];
1048
1049         if (!task)
1050                 return -ESRCH;
1051         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1052                                 audit_get_loginuid(task));
1053         put_task_struct(task);
1054         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1055 }
1056
1057 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1058                                    size_t count, loff_t *ppos)
1059 {
1060         struct inode * inode = file->f_path.dentry->d_inode;
1061         char *page, *tmp;
1062         ssize_t length;
1063         uid_t loginuid;
1064
1065         if (!capable(CAP_AUDIT_CONTROL))
1066                 return -EPERM;
1067
1068         if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
1069                 return -EPERM;
1070
1071         if (count >= PAGE_SIZE)
1072                 count = PAGE_SIZE - 1;
1073
1074         if (*ppos != 0) {
1075                 /* No partial writes. */
1076                 return -EINVAL;
1077         }
1078         page = (char*)__get_free_page(GFP_TEMPORARY);
1079         if (!page)
1080                 return -ENOMEM;
1081         length = -EFAULT;
1082         if (copy_from_user(page, buf, count))
1083                 goto out_free_page;
1084
1085         page[count] = '\0';
1086         loginuid = simple_strtoul(page, &tmp, 10);
1087         if (tmp == page) {
1088                 length = -EINVAL;
1089                 goto out_free_page;
1090
1091         }
1092         length = audit_set_loginuid(current, loginuid);
1093         if (likely(length == 0))
1094                 length = count;
1095
1096 out_free_page:
1097         free_page((unsigned long) page);
1098         return length;
1099 }
1100
1101 static const struct file_operations proc_loginuid_operations = {
1102         .read           = proc_loginuid_read,
1103         .write          = proc_loginuid_write,
1104 };
1105
1106 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1107                                   size_t count, loff_t *ppos)
1108 {
1109         struct inode * inode = file->f_path.dentry->d_inode;
1110         struct task_struct *task = get_proc_task(inode);
1111         ssize_t length;
1112         char tmpbuf[TMPBUFLEN];
1113
1114         if (!task)
1115                 return -ESRCH;
1116         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1117                                 audit_get_sessionid(task));
1118         put_task_struct(task);
1119         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1120 }
1121
1122 static const struct file_operations proc_sessionid_operations = {
1123         .read           = proc_sessionid_read,
1124 };
1125 #endif
1126
1127 #ifdef CONFIG_FAULT_INJECTION
1128 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1129                                       size_t count, loff_t *ppos)
1130 {
1131         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1132         char buffer[PROC_NUMBUF];
1133         size_t len;
1134         int make_it_fail;
1135
1136         if (!task)
1137                 return -ESRCH;
1138         make_it_fail = task->make_it_fail;
1139         put_task_struct(task);
1140
1141         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1142
1143         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1144 }
1145
1146 static ssize_t proc_fault_inject_write(struct file * file,
1147                         const char __user * buf, size_t count, loff_t *ppos)
1148 {
1149         struct task_struct *task;
1150         char buffer[PROC_NUMBUF], *end;
1151         int make_it_fail;
1152
1153         if (!capable(CAP_SYS_RESOURCE))
1154                 return -EPERM;
1155         memset(buffer, 0, sizeof(buffer));
1156         if (count > sizeof(buffer) - 1)
1157                 count = sizeof(buffer) - 1;
1158         if (copy_from_user(buffer, buf, count))
1159                 return -EFAULT;
1160         make_it_fail = simple_strtol(buffer, &end, 0);
1161         if (*end == '\n')
1162                 end++;
1163         task = get_proc_task(file->f_dentry->d_inode);
1164         if (!task)
1165                 return -ESRCH;
1166         task->make_it_fail = make_it_fail;
1167         put_task_struct(task);
1168         if (end - buffer == 0)
1169                 return -EIO;
1170         return end - buffer;
1171 }
1172
1173 static const struct file_operations proc_fault_inject_operations = {
1174         .read           = proc_fault_inject_read,
1175         .write          = proc_fault_inject_write,
1176 };
1177 #endif
1178
1179
1180 #ifdef CONFIG_SCHED_DEBUG
1181 /*
1182  * Print out various scheduling related per-task fields:
1183  */
1184 static int sched_show(struct seq_file *m, void *v)
1185 {
1186         struct inode *inode = m->private;
1187         struct task_struct *p;
1188
1189         WARN_ON(!inode);
1190
1191         p = get_proc_task(inode);
1192         if (!p)
1193                 return -ESRCH;
1194         proc_sched_show_task(p, m);
1195
1196         put_task_struct(p);
1197
1198         return 0;
1199 }
1200
1201 static ssize_t
1202 sched_write(struct file *file, const char __user *buf,
1203             size_t count, loff_t *offset)
1204 {
1205         struct inode *inode = file->f_path.dentry->d_inode;
1206         struct task_struct *p;
1207
1208         WARN_ON(!inode);
1209
1210         p = get_proc_task(inode);
1211         if (!p)
1212                 return -ESRCH;
1213         proc_sched_set_task(p);
1214
1215         put_task_struct(p);
1216
1217         return count;
1218 }
1219
1220 static int sched_open(struct inode *inode, struct file *filp)
1221 {
1222         int ret;
1223
1224         ret = single_open(filp, sched_show, NULL);
1225         if (!ret) {
1226                 struct seq_file *m = filp->private_data;
1227
1228                 m->private = inode;
1229         }
1230         return ret;
1231 }
1232
1233 static const struct file_operations proc_pid_sched_operations = {
1234         .open           = sched_open,
1235         .read           = seq_read,
1236         .write          = sched_write,
1237         .llseek         = seq_lseek,
1238         .release        = single_release,
1239 };
1240
1241 #endif
1242
1243 /*
1244  * We added or removed a vma mapping the executable. The vmas are only mapped
1245  * during exec and are not mapped with the mmap system call.
1246  * Callers must hold down_write() on the mm's mmap_sem for these
1247  */
1248 void added_exe_file_vma(struct mm_struct *mm)
1249 {
1250         mm->num_exe_file_vmas++;
1251 }
1252
1253 void removed_exe_file_vma(struct mm_struct *mm)
1254 {
1255         mm->num_exe_file_vmas--;
1256         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1257                 fput(mm->exe_file);
1258                 mm->exe_file = NULL;
1259         }
1260
1261 }
1262
1263 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1264 {
1265         if (new_exe_file)
1266                 get_file(new_exe_file);
1267         if (mm->exe_file)
1268                 fput(mm->exe_file);
1269         mm->exe_file = new_exe_file;
1270         mm->num_exe_file_vmas = 0;
1271 }
1272
1273 struct file *get_mm_exe_file(struct mm_struct *mm)
1274 {
1275         struct file *exe_file;
1276
1277         /* We need mmap_sem to protect against races with removal of
1278          * VM_EXECUTABLE vmas */
1279         down_read(&mm->mmap_sem);
1280         exe_file = mm->exe_file;
1281         if (exe_file)
1282                 get_file(exe_file);
1283         up_read(&mm->mmap_sem);
1284         return exe_file;
1285 }
1286
1287 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1288 {
1289         /* It's safe to write the exe_file pointer without exe_file_lock because
1290          * this is called during fork when the task is not yet in /proc */
1291         newmm->exe_file = get_mm_exe_file(oldmm);
1292 }
1293
1294 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1295 {
1296         struct task_struct *task;
1297         struct mm_struct *mm;
1298         struct file *exe_file;
1299
1300         task = get_proc_task(inode);
1301         if (!task)
1302                 return -ENOENT;
1303         mm = get_task_mm(task);
1304         put_task_struct(task);
1305         if (!mm)
1306                 return -ENOENT;
1307         exe_file = get_mm_exe_file(mm);
1308         mmput(mm);
1309         if (exe_file) {
1310                 *exe_path = exe_file->f_path;
1311                 path_get(&exe_file->f_path);
1312                 fput(exe_file);
1313                 return 0;
1314         } else
1315                 return -ENOENT;
1316 }
1317
1318 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1319 {
1320         struct inode *inode = dentry->d_inode;
1321         int error = -EACCES;
1322
1323         /* We don't need a base pointer in the /proc filesystem */
1324         path_put(&nd->path);
1325
1326         /* Are we allowed to snoop on the tasks file descriptors? */
1327         if (!proc_fd_access_allowed(inode))
1328                 goto out;
1329
1330         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1331         nd->last_type = LAST_BIND;
1332 out:
1333         return ERR_PTR(error);
1334 }
1335
1336 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1337 {
1338         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1339         char *pathname;
1340         int len;
1341
1342         if (!tmp)
1343                 return -ENOMEM;
1344
1345         pathname = d_path(path, tmp, PAGE_SIZE);
1346         len = PTR_ERR(pathname);
1347         if (IS_ERR(pathname))
1348                 goto out;
1349         len = tmp + PAGE_SIZE - 1 - pathname;
1350
1351         if (len > buflen)
1352                 len = buflen;
1353         if (copy_to_user(buffer, pathname, len))
1354                 len = -EFAULT;
1355  out:
1356         free_page((unsigned long)tmp);
1357         return len;
1358 }
1359
1360 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1361 {
1362         int error = -EACCES;
1363         struct inode *inode = dentry->d_inode;
1364         struct path path;
1365
1366         /* Are we allowed to snoop on the tasks file descriptors? */
1367         if (!proc_fd_access_allowed(inode))
1368                 goto out;
1369
1370         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1371         if (error)
1372                 goto out;
1373
1374         error = do_proc_readlink(&path, buffer, buflen);
1375         path_put(&path);
1376 out:
1377         return error;
1378 }
1379
1380 static const struct inode_operations proc_pid_link_inode_operations = {
1381         .readlink       = proc_pid_readlink,
1382         .follow_link    = proc_pid_follow_link,
1383         .setattr        = proc_setattr,
1384 };
1385
1386
1387 /* building an inode */
1388
1389 static int task_dumpable(struct task_struct *task)
1390 {
1391         int dumpable = 0;
1392         struct mm_struct *mm;
1393
1394         task_lock(task);
1395         mm = task->mm;
1396         if (mm)
1397                 dumpable = get_dumpable(mm);
1398         task_unlock(task);
1399         if(dumpable == 1)
1400                 return 1;
1401         return 0;
1402 }
1403
1404
1405 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1406 {
1407         struct inode * inode;
1408         struct proc_inode *ei;
1409         const struct cred *cred;
1410
1411         /* We need a new inode */
1412
1413         inode = new_inode(sb);
1414         if (!inode)
1415                 goto out;
1416
1417         /* Common stuff */
1418         ei = PROC_I(inode);
1419         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1420         inode->i_op = &proc_def_inode_operations;
1421
1422         /*
1423          * grab the reference to task.
1424          */
1425         ei->pid = get_task_pid(task, PIDTYPE_PID);
1426         if (!ei->pid)
1427                 goto out_unlock;
1428
1429         if (task_dumpable(task)) {
1430                 rcu_read_lock();
1431                 cred = __task_cred(task);
1432                 inode->i_uid = cred->euid;
1433                 inode->i_gid = cred->egid;
1434                 rcu_read_unlock();
1435         }
1436         security_task_to_inode(task, inode);
1437
1438 out:
1439         return inode;
1440
1441 out_unlock:
1442         iput(inode);
1443         return NULL;
1444 }
1445
1446 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1447 {
1448         struct inode *inode = dentry->d_inode;
1449         struct task_struct *task;
1450         const struct cred *cred;
1451
1452         generic_fillattr(inode, stat);
1453
1454         rcu_read_lock();
1455         stat->uid = 0;
1456         stat->gid = 0;
1457         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1458         if (task) {
1459                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1460                     task_dumpable(task)) {
1461                         cred = __task_cred(task);
1462                         stat->uid = cred->euid;
1463                         stat->gid = cred->egid;
1464                 }
1465         }
1466         rcu_read_unlock();
1467         return 0;
1468 }
1469
1470 /* dentry stuff */
1471
1472 /*
1473  *      Exceptional case: normally we are not allowed to unhash a busy
1474  * directory. In this case, however, we can do it - no aliasing problems
1475  * due to the way we treat inodes.
1476  *
1477  * Rewrite the inode's ownerships here because the owning task may have
1478  * performed a setuid(), etc.
1479  *
1480  * Before the /proc/pid/status file was created the only way to read
1481  * the effective uid of a /process was to stat /proc/pid.  Reading
1482  * /proc/pid/status is slow enough that procps and other packages
1483  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1484  * made this apply to all per process world readable and executable
1485  * directories.
1486  */
1487 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1488 {
1489         struct inode *inode = dentry->d_inode;
1490         struct task_struct *task = get_proc_task(inode);
1491         const struct cred *cred;
1492
1493         if (task) {
1494                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1495                     task_dumpable(task)) {
1496                         rcu_read_lock();
1497                         cred = __task_cred(task);
1498                         inode->i_uid = cred->euid;
1499                         inode->i_gid = cred->egid;
1500                         rcu_read_unlock();
1501                 } else {
1502                         inode->i_uid = 0;
1503                         inode->i_gid = 0;
1504                 }
1505                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1506                 security_task_to_inode(task, inode);
1507                 put_task_struct(task);
1508                 return 1;
1509         }
1510         d_drop(dentry);
1511         return 0;
1512 }
1513
1514 static int pid_delete_dentry(struct dentry * dentry)
1515 {
1516         /* Is the task we represent dead?
1517          * If so, then don't put the dentry on the lru list,
1518          * kill it immediately.
1519          */
1520         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1521 }
1522
1523 static struct dentry_operations pid_dentry_operations =
1524 {
1525         .d_revalidate   = pid_revalidate,
1526         .d_delete       = pid_delete_dentry,
1527 };
1528
1529 /* Lookups */
1530
1531 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1532                                 struct task_struct *, const void *);
1533
1534 /*
1535  * Fill a directory entry.
1536  *
1537  * If possible create the dcache entry and derive our inode number and
1538  * file type from dcache entry.
1539  *
1540  * Since all of the proc inode numbers are dynamically generated, the inode
1541  * numbers do not exist until the inode is cache.  This means creating the
1542  * the dcache entry in readdir is necessary to keep the inode numbers
1543  * reported by readdir in sync with the inode numbers reported
1544  * by stat.
1545  */
1546 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1547         char *name, int len,
1548         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1549 {
1550         struct dentry *child, *dir = filp->f_path.dentry;
1551         struct inode *inode;
1552         struct qstr qname;
1553         ino_t ino = 0;
1554         unsigned type = DT_UNKNOWN;
1555
1556         qname.name = name;
1557         qname.len  = len;
1558         qname.hash = full_name_hash(name, len);
1559
1560         child = d_lookup(dir, &qname);
1561         if (!child) {
1562                 struct dentry *new;
1563                 new = d_alloc(dir, &qname);
1564                 if (new) {
1565                         child = instantiate(dir->d_inode, new, task, ptr);
1566                         if (child)
1567                                 dput(new);
1568                         else
1569                                 child = new;
1570                 }
1571         }
1572         if (!child || IS_ERR(child) || !child->d_inode)
1573                 goto end_instantiate;
1574         inode = child->d_inode;
1575         if (inode) {
1576                 ino = inode->i_ino;
1577                 type = inode->i_mode >> 12;
1578         }
1579         dput(child);
1580 end_instantiate:
1581         if (!ino)
1582                 ino = find_inode_number(dir, &qname);
1583         if (!ino)
1584                 ino = 1;
1585         return filldir(dirent, name, len, filp->f_pos, ino, type);
1586 }
1587
1588 static unsigned name_to_int(struct dentry *dentry)
1589 {
1590         const char *name = dentry->d_name.name;
1591         int len = dentry->d_name.len;
1592         unsigned n = 0;
1593
1594         if (len > 1 && *name == '0')
1595                 goto out;
1596         while (len-- > 0) {
1597                 unsigned c = *name++ - '0';
1598                 if (c > 9)
1599                         goto out;
1600                 if (n >= (~0U-9)/10)
1601                         goto out;
1602                 n *= 10;
1603                 n += c;
1604         }
1605         return n;
1606 out:
1607         return ~0U;
1608 }
1609
1610 #define PROC_FDINFO_MAX 64
1611
1612 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1613 {
1614         struct task_struct *task = get_proc_task(inode);
1615         struct files_struct *files = NULL;
1616         struct file *file;
1617         int fd = proc_fd(inode);
1618
1619         if (task) {
1620                 files = get_files_struct(task);
1621                 put_task_struct(task);
1622         }
1623         if (files) {
1624                 /*
1625                  * We are not taking a ref to the file structure, so we must
1626                  * hold ->file_lock.
1627                  */
1628                 spin_lock(&files->file_lock);
1629                 file = fcheck_files(files, fd);
1630                 if (file) {
1631                         if (path) {
1632                                 *path = file->f_path;
1633                                 path_get(&file->f_path);
1634                         }
1635                         if (info)
1636                                 snprintf(info, PROC_FDINFO_MAX,
1637                                          "pos:\t%lli\n"
1638                                          "flags:\t0%o\n",
1639                                          (long long) file->f_pos,
1640                                          file->f_flags);
1641                         spin_unlock(&files->file_lock);
1642                         put_files_struct(files);
1643                         return 0;
1644                 }
1645                 spin_unlock(&files->file_lock);
1646                 put_files_struct(files);
1647         }
1648         return -ENOENT;
1649 }
1650
1651 static int proc_fd_link(struct inode *inode, struct path *path)
1652 {
1653         return proc_fd_info(inode, path, NULL);
1654 }
1655
1656 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1657 {
1658         struct inode *inode = dentry->d_inode;
1659         struct task_struct *task = get_proc_task(inode);
1660         int fd = proc_fd(inode);
1661         struct files_struct *files;
1662         const struct cred *cred;
1663
1664         if (task) {
1665                 files = get_files_struct(task);
1666                 if (files) {
1667                         rcu_read_lock();
1668                         if (fcheck_files(files, fd)) {
1669                                 rcu_read_unlock();
1670                                 put_files_struct(files);
1671                                 if (task_dumpable(task)) {
1672                                         rcu_read_lock();
1673                                         cred = __task_cred(task);
1674                                         inode->i_uid = cred->euid;
1675                                         inode->i_gid = cred->egid;
1676                                         rcu_read_unlock();
1677                                 } else {
1678                                         inode->i_uid = 0;
1679                                         inode->i_gid = 0;
1680                                 }
1681                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1682                                 security_task_to_inode(task, inode);
1683                                 put_task_struct(task);
1684                                 return 1;
1685                         }
1686                         rcu_read_unlock();
1687                         put_files_struct(files);
1688                 }
1689                 put_task_struct(task);
1690         }
1691         d_drop(dentry);
1692         return 0;
1693 }
1694
1695 static struct dentry_operations tid_fd_dentry_operations =
1696 {
1697         .d_revalidate   = tid_fd_revalidate,
1698         .d_delete       = pid_delete_dentry,
1699 };
1700
1701 static struct dentry *proc_fd_instantiate(struct inode *dir,
1702         struct dentry *dentry, struct task_struct *task, const void *ptr)
1703 {
1704         unsigned fd = *(const unsigned *)ptr;
1705         struct file *file;
1706         struct files_struct *files;
1707         struct inode *inode;
1708         struct proc_inode *ei;
1709         struct dentry *error = ERR_PTR(-ENOENT);
1710
1711         inode = proc_pid_make_inode(dir->i_sb, task);
1712         if (!inode)
1713                 goto out;
1714         ei = PROC_I(inode);
1715         ei->fd = fd;
1716         files = get_files_struct(task);
1717         if (!files)
1718                 goto out_iput;
1719         inode->i_mode = S_IFLNK;
1720
1721         /*
1722          * We are not taking a ref to the file structure, so we must
1723          * hold ->file_lock.
1724          */
1725         spin_lock(&files->file_lock);
1726         file = fcheck_files(files, fd);
1727         if (!file)
1728                 goto out_unlock;
1729         if (file->f_mode & FMODE_READ)
1730                 inode->i_mode |= S_IRUSR | S_IXUSR;
1731         if (file->f_mode & FMODE_WRITE)
1732                 inode->i_mode |= S_IWUSR | S_IXUSR;
1733         spin_unlock(&files->file_lock);
1734         put_files_struct(files);
1735
1736         inode->i_op = &proc_pid_link_inode_operations;
1737         inode->i_size = 64;
1738         ei->op.proc_get_link = proc_fd_link;
1739         dentry->d_op = &tid_fd_dentry_operations;
1740         d_add(dentry, inode);
1741         /* Close the race of the process dying before we return the dentry */
1742         if (tid_fd_revalidate(dentry, NULL))
1743                 error = NULL;
1744
1745  out:
1746         return error;
1747 out_unlock:
1748         spin_unlock(&files->file_lock);
1749         put_files_struct(files);
1750 out_iput:
1751         iput(inode);
1752         goto out;
1753 }
1754
1755 static struct dentry *proc_lookupfd_common(struct inode *dir,
1756                                            struct dentry *dentry,
1757                                            instantiate_t instantiate)
1758 {
1759         struct task_struct *task = get_proc_task(dir);
1760         unsigned fd = name_to_int(dentry);
1761         struct dentry *result = ERR_PTR(-ENOENT);
1762
1763         if (!task)
1764                 goto out_no_task;
1765         if (fd == ~0U)
1766                 goto out;
1767
1768         result = instantiate(dir, dentry, task, &fd);
1769 out:
1770         put_task_struct(task);
1771 out_no_task:
1772         return result;
1773 }
1774
1775 static int proc_readfd_common(struct file * filp, void * dirent,
1776                               filldir_t filldir, instantiate_t instantiate)
1777 {
1778         struct dentry *dentry = filp->f_path.dentry;
1779         struct inode *inode = dentry->d_inode;
1780         struct task_struct *p = get_proc_task(inode);
1781         unsigned int fd, ino;
1782         int retval;
1783         struct files_struct * files;
1784
1785         retval = -ENOENT;
1786         if (!p)
1787                 goto out_no_task;
1788         retval = 0;
1789
1790         fd = filp->f_pos;
1791         switch (fd) {
1792                 case 0:
1793                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1794                                 goto out;
1795                         filp->f_pos++;
1796                 case 1:
1797                         ino = parent_ino(dentry);
1798                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1799                                 goto out;
1800                         filp->f_pos++;
1801                 default:
1802                         files = get_files_struct(p);
1803                         if (!files)
1804                                 goto out;
1805                         rcu_read_lock();
1806                         for (fd = filp->f_pos-2;
1807                              fd < files_fdtable(files)->max_fds;
1808                              fd++, filp->f_pos++) {
1809                                 char name[PROC_NUMBUF];
1810                                 int len;
1811
1812                                 if (!fcheck_files(files, fd))
1813                                         continue;
1814                                 rcu_read_unlock();
1815
1816                                 len = snprintf(name, sizeof(name), "%d", fd);
1817                                 if (proc_fill_cache(filp, dirent, filldir,
1818                                                     name, len, instantiate,
1819                                                     p, &fd) < 0) {
1820                                         rcu_read_lock();
1821                                         break;
1822                                 }
1823                                 rcu_read_lock();
1824                         }
1825                         rcu_read_unlock();
1826                         put_files_struct(files);
1827         }
1828 out:
1829         put_task_struct(p);
1830 out_no_task:
1831         return retval;
1832 }
1833
1834 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1835                                     struct nameidata *nd)
1836 {
1837         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1838 }
1839
1840 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1841 {
1842         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1843 }
1844
1845 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1846                                       size_t len, loff_t *ppos)
1847 {
1848         char tmp[PROC_FDINFO_MAX];
1849         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1850         if (!err)
1851                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1852         return err;
1853 }
1854
1855 static const struct file_operations proc_fdinfo_file_operations = {
1856         .open           = nonseekable_open,
1857         .read           = proc_fdinfo_read,
1858 };
1859
1860 static const struct file_operations proc_fd_operations = {
1861         .read           = generic_read_dir,
1862         .readdir        = proc_readfd,
1863 };
1864
1865 /*
1866  * /proc/pid/fd needs a special permission handler so that a process can still
1867  * access /proc/self/fd after it has executed a setuid().
1868  */
1869 static int proc_fd_permission(struct inode *inode, int mask)
1870 {
1871         int rv;
1872
1873         rv = generic_permission(inode, mask, NULL);
1874         if (rv == 0)
1875                 return 0;
1876         if (task_pid(current) == proc_pid(inode))
1877                 rv = 0;
1878         return rv;
1879 }
1880
1881 /*
1882  * proc directories can do almost nothing..
1883  */
1884 static const struct inode_operations proc_fd_inode_operations = {
1885         .lookup         = proc_lookupfd,
1886         .permission     = proc_fd_permission,
1887         .setattr        = proc_setattr,
1888 };
1889
1890 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1891         struct dentry *dentry, struct task_struct *task, const void *ptr)
1892 {
1893         unsigned fd = *(unsigned *)ptr;
1894         struct inode *inode;
1895         struct proc_inode *ei;
1896         struct dentry *error = ERR_PTR(-ENOENT);
1897
1898         inode = proc_pid_make_inode(dir->i_sb, task);
1899         if (!inode)
1900                 goto out;
1901         ei = PROC_I(inode);
1902         ei->fd = fd;
1903         inode->i_mode = S_IFREG | S_IRUSR;
1904         inode->i_fop = &proc_fdinfo_file_operations;
1905         dentry->d_op = &tid_fd_dentry_operations;
1906         d_add(dentry, inode);
1907         /* Close the race of the process dying before we return the dentry */
1908         if (tid_fd_revalidate(dentry, NULL))
1909                 error = NULL;
1910
1911  out:
1912         return error;
1913 }
1914
1915 static struct dentry *proc_lookupfdinfo(struct inode *dir,
1916                                         struct dentry *dentry,
1917                                         struct nameidata *nd)
1918 {
1919         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
1920 }
1921
1922 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
1923 {
1924         return proc_readfd_common(filp, dirent, filldir,
1925                                   proc_fdinfo_instantiate);
1926 }
1927
1928 static const struct file_operations proc_fdinfo_operations = {
1929         .read           = generic_read_dir,
1930         .readdir        = proc_readfdinfo,
1931 };
1932
1933 /*
1934  * proc directories can do almost nothing..
1935  */
1936 static const struct inode_operations proc_fdinfo_inode_operations = {
1937         .lookup         = proc_lookupfdinfo,
1938         .setattr        = proc_setattr,
1939 };
1940
1941
1942 static struct dentry *proc_pident_instantiate(struct inode *dir,
1943         struct dentry *dentry, struct task_struct *task, const void *ptr)
1944 {
1945         const struct pid_entry *p = ptr;
1946         struct inode *inode;
1947         struct proc_inode *ei;
1948         struct dentry *error = ERR_PTR(-EINVAL);
1949
1950         inode = proc_pid_make_inode(dir->i_sb, task);
1951         if (!inode)
1952                 goto out;
1953
1954         ei = PROC_I(inode);
1955         inode->i_mode = p->mode;
1956         if (S_ISDIR(inode->i_mode))
1957                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
1958         if (p->iop)
1959                 inode->i_op = p->iop;
1960         if (p->fop)
1961                 inode->i_fop = p->fop;
1962         ei->op = p->op;
1963         dentry->d_op = &pid_dentry_operations;
1964         d_add(dentry, inode);
1965         /* Close the race of the process dying before we return the dentry */
1966         if (pid_revalidate(dentry, NULL))
1967                 error = NULL;
1968 out:
1969         return error;
1970 }
1971
1972 static struct dentry *proc_pident_lookup(struct inode *dir,
1973                                          struct dentry *dentry,
1974                                          const struct pid_entry *ents,
1975                                          unsigned int nents)
1976 {
1977         struct inode *inode;
1978         struct dentry *error;
1979         struct task_struct *task = get_proc_task(dir);
1980         const struct pid_entry *p, *last;
1981
1982         error = ERR_PTR(-ENOENT);
1983         inode = NULL;
1984
1985         if (!task)
1986                 goto out_no_task;
1987
1988         /*
1989          * Yes, it does not scale. And it should not. Don't add
1990          * new entries into /proc/<tgid>/ without very good reasons.
1991          */
1992         last = &ents[nents - 1];
1993         for (p = ents; p <= last; p++) {
1994                 if (p->len != dentry->d_name.len)
1995                         continue;
1996                 if (!memcmp(dentry->d_name.name, p->name, p->len))
1997                         break;
1998         }
1999         if (p > last)
2000                 goto out;
2001
2002         error = proc_pident_instantiate(dir, dentry, task, p);
2003 out:
2004         put_task_struct(task);
2005 out_no_task:
2006         return error;
2007 }
2008
2009 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2010         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2011 {
2012         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2013                                 proc_pident_instantiate, task, p);
2014 }
2015
2016 static int proc_pident_readdir(struct file *filp,
2017                 void *dirent, filldir_t filldir,
2018                 const struct pid_entry *ents, unsigned int nents)
2019 {
2020         int i;
2021         struct dentry *dentry = filp->f_path.dentry;
2022         struct inode *inode = dentry->d_inode;
2023         struct task_struct *task = get_proc_task(inode);
2024         const struct pid_entry *p, *last;
2025         ino_t ino;
2026         int ret;
2027
2028         ret = -ENOENT;
2029         if (!task)
2030                 goto out_no_task;
2031
2032         ret = 0;
2033         i = filp->f_pos;
2034         switch (i) {
2035         case 0:
2036                 ino = inode->i_ino;
2037                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2038                         goto out;
2039                 i++;
2040                 filp->f_pos++;
2041                 /* fall through */
2042         case 1:
2043                 ino = parent_ino(dentry);
2044                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2045                         goto out;
2046                 i++;
2047                 filp->f_pos++;
2048                 /* fall through */
2049         default:
2050                 i -= 2;
2051                 if (i >= nents) {
2052                         ret = 1;
2053                         goto out;
2054                 }
2055                 p = ents + i;
2056                 last = &ents[nents - 1];
2057                 while (p <= last) {
2058                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2059                                 goto out;
2060                         filp->f_pos++;
2061                         p++;
2062                 }
2063         }
2064
2065         ret = 1;
2066 out:
2067         put_task_struct(task);
2068 out_no_task:
2069         return ret;
2070 }
2071
2072 #ifdef CONFIG_SECURITY
2073 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2074                                   size_t count, loff_t *ppos)
2075 {
2076         struct inode * inode = file->f_path.dentry->d_inode;
2077         char *p = NULL;
2078         ssize_t length;
2079         struct task_struct *task = get_proc_task(inode);
2080
2081         if (!task)
2082                 return -ESRCH;
2083
2084         length = security_getprocattr(task,
2085                                       (char*)file->f_path.dentry->d_name.name,
2086                                       &p);
2087         put_task_struct(task);
2088         if (length > 0)
2089                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2090         kfree(p);
2091         return length;
2092 }
2093
2094 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2095                                    size_t count, loff_t *ppos)
2096 {
2097         struct inode * inode = file->f_path.dentry->d_inode;
2098         char *page;
2099         ssize_t length;
2100         struct task_struct *task = get_proc_task(inode);
2101
2102         length = -ESRCH;
2103         if (!task)
2104                 goto out_no_task;
2105         if (count > PAGE_SIZE)
2106                 count = PAGE_SIZE;
2107
2108         /* No partial writes. */
2109         length = -EINVAL;
2110         if (*ppos != 0)
2111                 goto out;
2112
2113         length = -ENOMEM;
2114         page = (char*)__get_free_page(GFP_TEMPORARY);
2115         if (!page)
2116                 goto out;
2117
2118         length = -EFAULT;
2119         if (copy_from_user(page, buf, count))
2120                 goto out_free;
2121
2122         length = security_setprocattr(task,
2123                                       (char*)file->f_path.dentry->d_name.name,
2124                                       (void*)page, count);
2125 out_free:
2126         free_page((unsigned long) page);
2127 out:
2128         put_task_struct(task);
2129 out_no_task:
2130         return length;
2131 }
2132
2133 static const struct file_operations proc_pid_attr_operations = {
2134         .read           = proc_pid_attr_read,
2135         .write          = proc_pid_attr_write,
2136 };
2137
2138 static const struct pid_entry attr_dir_stuff[] = {
2139         REG("current",    S_IRUGO|S_IWUGO, pid_attr),
2140         REG("prev",       S_IRUGO,         pid_attr),
2141         REG("exec",       S_IRUGO|S_IWUGO, pid_attr),
2142         REG("fscreate",   S_IRUGO|S_IWUGO, pid_attr),
2143         REG("keycreate",  S_IRUGO|S_IWUGO, pid_attr),
2144         REG("sockcreate", S_IRUGO|S_IWUGO, pid_attr),
2145 };
2146
2147 static int proc_attr_dir_readdir(struct file * filp,
2148                              void * dirent, filldir_t filldir)
2149 {
2150         return proc_pident_readdir(filp,dirent,filldir,
2151                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2152 }
2153
2154 static const struct file_operations proc_attr_dir_operations = {
2155         .read           = generic_read_dir,
2156         .readdir        = proc_attr_dir_readdir,
2157 };
2158
2159 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2160                                 struct dentry *dentry, struct nameidata *nd)
2161 {
2162         return proc_pident_lookup(dir, dentry,
2163                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2164 }
2165
2166 static const struct inode_operations proc_attr_dir_inode_operations = {
2167         .lookup         = proc_attr_dir_lookup,
2168         .getattr        = pid_getattr,
2169         .setattr        = proc_setattr,
2170 };
2171
2172 #endif
2173
2174 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2175 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2176                                          size_t count, loff_t *ppos)
2177 {
2178         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2179         struct mm_struct *mm;
2180         char buffer[PROC_NUMBUF];
2181         size_t len;
2182         int ret;
2183
2184         if (!task)
2185                 return -ESRCH;
2186
2187         ret = 0;
2188         mm = get_task_mm(task);
2189         if (mm) {
2190                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2191                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2192                                 MMF_DUMP_FILTER_SHIFT));
2193                 mmput(mm);
2194                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2195         }
2196
2197         put_task_struct(task);
2198
2199         return ret;
2200 }
2201
2202 static ssize_t proc_coredump_filter_write(struct file *file,
2203                                           const char __user *buf,
2204                                           size_t count,
2205                                           loff_t *ppos)
2206 {
2207         struct task_struct *task;
2208         struct mm_struct *mm;
2209         char buffer[PROC_NUMBUF], *end;
2210         unsigned int val;
2211         int ret;
2212         int i;
2213         unsigned long mask;
2214
2215         ret = -EFAULT;
2216         memset(buffer, 0, sizeof(buffer));
2217         if (count > sizeof(buffer) - 1)
2218                 count = sizeof(buffer) - 1;
2219         if (copy_from_user(buffer, buf, count))
2220                 goto out_no_task;
2221
2222         ret = -EINVAL;
2223         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2224         if (*end == '\n')
2225                 end++;
2226         if (end - buffer == 0)
2227                 goto out_no_task;
2228
2229         ret = -ESRCH;
2230         task = get_proc_task(file->f_dentry->d_inode);
2231         if (!task)
2232                 goto out_no_task;
2233
2234         ret = end - buffer;
2235         mm = get_task_mm(task);
2236         if (!mm)
2237                 goto out_no_mm;
2238
2239         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2240                 if (val & mask)
2241                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2242                 else
2243                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2244         }
2245
2246         mmput(mm);
2247  out_no_mm:
2248         put_task_struct(task);
2249  out_no_task:
2250         return ret;
2251 }
2252
2253 static const struct file_operations proc_coredump_filter_operations = {
2254         .read           = proc_coredump_filter_read,
2255         .write          = proc_coredump_filter_write,
2256 };
2257 #endif
2258
2259 /*
2260  * /proc/self:
2261  */
2262 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2263                               int buflen)
2264 {
2265         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2266         pid_t tgid = task_tgid_nr_ns(current, ns);
2267         char tmp[PROC_NUMBUF];
2268         if (!tgid)
2269                 return -ENOENT;
2270         sprintf(tmp, "%d", tgid);
2271         return vfs_readlink(dentry,buffer,buflen,tmp);
2272 }
2273
2274 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2275 {
2276         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2277         pid_t tgid = task_tgid_nr_ns(current, ns);
2278         char tmp[PROC_NUMBUF];
2279         if (!tgid)
2280                 return ERR_PTR(-ENOENT);
2281         sprintf(tmp, "%d", task_tgid_nr_ns(current, ns));
2282         return ERR_PTR(vfs_follow_link(nd,tmp));
2283 }
2284
2285 static const struct inode_operations proc_self_inode_operations = {
2286         .readlink       = proc_self_readlink,
2287         .follow_link    = proc_self_follow_link,
2288 };
2289
2290 /*
2291  * proc base
2292  *
2293  * These are the directory entries in the root directory of /proc
2294  * that properly belong to the /proc filesystem, as they describe
2295  * describe something that is process related.
2296  */
2297 static const struct pid_entry proc_base_stuff[] = {
2298         NOD("self", S_IFLNK|S_IRWXUGO,
2299                 &proc_self_inode_operations, NULL, {}),
2300 };
2301
2302 /*
2303  *      Exceptional case: normally we are not allowed to unhash a busy
2304  * directory. In this case, however, we can do it - no aliasing problems
2305  * due to the way we treat inodes.
2306  */
2307 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2308 {
2309         struct inode *inode = dentry->d_inode;
2310         struct task_struct *task = get_proc_task(inode);
2311         if (task) {
2312                 put_task_struct(task);
2313                 return 1;
2314         }
2315         d_drop(dentry);
2316         return 0;
2317 }
2318
2319 static struct dentry_operations proc_base_dentry_operations =
2320 {
2321         .d_revalidate   = proc_base_revalidate,
2322         .d_delete       = pid_delete_dentry,
2323 };
2324
2325 static struct dentry *proc_base_instantiate(struct inode *dir,
2326         struct dentry *dentry, struct task_struct *task, const void *ptr)
2327 {
2328         const struct pid_entry *p = ptr;
2329         struct inode *inode;
2330         struct proc_inode *ei;
2331         struct dentry *error = ERR_PTR(-EINVAL);
2332
2333         /* Allocate the inode */
2334         error = ERR_PTR(-ENOMEM);
2335         inode = new_inode(dir->i_sb);
2336         if (!inode)
2337                 goto out;
2338
2339         /* Initialize the inode */
2340         ei = PROC_I(inode);
2341         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2342
2343         /*
2344          * grab the reference to the task.
2345          */
2346         ei->pid = get_task_pid(task, PIDTYPE_PID);
2347         if (!ei->pid)
2348                 goto out_iput;
2349
2350         inode->i_mode = p->mode;
2351         if (S_ISDIR(inode->i_mode))
2352                 inode->i_nlink = 2;
2353         if (S_ISLNK(inode->i_mode))
2354                 inode->i_size = 64;
2355         if (p->iop)
2356                 inode->i_op = p->iop;
2357         if (p->fop)
2358                 inode->i_fop = p->fop;
2359         ei->op = p->op;
2360         dentry->d_op = &proc_base_dentry_operations;
2361         d_add(dentry, inode);
2362         error = NULL;
2363 out:
2364         return error;
2365 out_iput:
2366         iput(inode);
2367         goto out;
2368 }
2369
2370 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2371 {
2372         struct dentry *error;
2373         struct task_struct *task = get_proc_task(dir);
2374         const struct pid_entry *p, *last;
2375
2376         error = ERR_PTR(-ENOENT);
2377
2378         if (!task)
2379                 goto out_no_task;
2380
2381         /* Lookup the directory entry */
2382         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2383         for (p = proc_base_stuff; p <= last; p++) {
2384                 if (p->len != dentry->d_name.len)
2385                         continue;
2386                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2387                         break;
2388         }
2389         if (p > last)
2390                 goto out;
2391
2392         error = proc_base_instantiate(dir, dentry, task, p);
2393
2394 out:
2395         put_task_struct(task);
2396 out_no_task:
2397         return error;
2398 }
2399
2400 static int proc_base_fill_cache(struct file *filp, void *dirent,
2401         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2402 {
2403         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2404                                 proc_base_instantiate, task, p);
2405 }
2406
2407 #ifdef CONFIG_TASK_IO_ACCOUNTING
2408 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2409 {
2410         struct task_io_accounting acct = task->ioac;
2411         unsigned long flags;
2412
2413         if (whole && lock_task_sighand(task, &flags)) {
2414                 struct task_struct *t = task;
2415
2416                 task_io_accounting_add(&acct, &task->signal->ioac);
2417                 while_each_thread(task, t)
2418                         task_io_accounting_add(&acct, &t->ioac);
2419
2420                 unlock_task_sighand(task, &flags);
2421         }
2422         return sprintf(buffer,
2423                         "rchar: %llu\n"
2424                         "wchar: %llu\n"
2425                         "syscr: %llu\n"
2426                         "syscw: %llu\n"
2427                         "read_bytes: %llu\n"
2428                         "write_bytes: %llu\n"
2429                         "cancelled_write_bytes: %llu\n",
2430                         (unsigned long long)acct.rchar,
2431                         (unsigned long long)acct.wchar,
2432                         (unsigned long long)acct.syscr,
2433                         (unsigned long long)acct.syscw,
2434                         (unsigned long long)acct.read_bytes,
2435                         (unsigned long long)acct.write_bytes,
2436                         (unsigned long long)acct.cancelled_write_bytes);
2437 }
2438
2439 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2440 {
2441         return do_io_accounting(task, buffer, 0);
2442 }
2443
2444 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2445 {
2446         return do_io_accounting(task, buffer, 1);
2447 }
2448 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2449
2450 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2451                                 struct pid *pid, struct task_struct *task)
2452 {
2453         seq_printf(m, "%08x\n", task->personality);
2454         return 0;
2455 }
2456
2457 /*
2458  * Thread groups
2459  */
2460 static const struct file_operations proc_task_operations;
2461 static const struct inode_operations proc_task_inode_operations;
2462
2463 static const struct pid_entry tgid_base_stuff[] = {
2464         DIR("task",       S_IRUGO|S_IXUGO, task),
2465         DIR("fd",         S_IRUSR|S_IXUSR, fd),
2466         DIR("fdinfo",     S_IRUSR|S_IXUSR, fdinfo),
2467 #ifdef CONFIG_NET
2468         DIR("net",        S_IRUGO|S_IXUGO, net),
2469 #endif
2470         REG("environ",    S_IRUSR, environ),
2471         INF("auxv",       S_IRUSR, pid_auxv),
2472         ONE("status",     S_IRUGO, pid_status),
2473         ONE("personality", S_IRUSR, pid_personality),
2474         INF("limits",     S_IRUSR, pid_limits),
2475 #ifdef CONFIG_SCHED_DEBUG
2476         REG("sched",      S_IRUGO|S_IWUSR, pid_sched),
2477 #endif
2478 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2479         INF("syscall",    S_IRUSR, pid_syscall),
2480 #endif
2481         INF("cmdline",    S_IRUGO, pid_cmdline),
2482         ONE("stat",       S_IRUGO, tgid_stat),
2483         ONE("statm",      S_IRUGO, pid_statm),
2484         REG("maps",       S_IRUGO, maps),
2485 #ifdef CONFIG_NUMA
2486         REG("numa_maps",  S_IRUGO, numa_maps),
2487 #endif
2488         REG("mem",        S_IRUSR|S_IWUSR, mem),
2489         LNK("cwd",        cwd),
2490         LNK("root",       root),
2491         LNK("exe",        exe),
2492         REG("mounts",     S_IRUGO, mounts),
2493         REG("mountinfo",  S_IRUGO, mountinfo),
2494         REG("mountstats", S_IRUSR, mountstats),
2495 #ifdef CONFIG_PROC_PAGE_MONITOR
2496         REG("clear_refs", S_IWUSR, clear_refs),
2497         REG("smaps",      S_IRUGO, smaps),
2498         REG("pagemap",    S_IRUSR, pagemap),
2499 #endif
2500 #ifdef CONFIG_SECURITY
2501         DIR("attr",       S_IRUGO|S_IXUGO, attr_dir),
2502 #endif
2503 #ifdef CONFIG_KALLSYMS
2504         INF("wchan",      S_IRUGO, pid_wchan),
2505 #endif
2506 #ifdef CONFIG_SCHEDSTATS
2507         INF("schedstat",  S_IRUGO, pid_schedstat),
2508 #endif
2509 #ifdef CONFIG_LATENCYTOP
2510         REG("latency",  S_IRUGO, lstats),
2511 #endif
2512 #ifdef CONFIG_PROC_PID_CPUSET
2513         REG("cpuset",     S_IRUGO, cpuset),
2514 #endif
2515 #ifdef CONFIG_CGROUPS
2516         REG("cgroup",  S_IRUGO, cgroup),
2517 #endif
2518         INF("oom_score",  S_IRUGO, oom_score),
2519         REG("oom_adj",    S_IRUGO|S_IWUSR, oom_adjust),
2520 #ifdef CONFIG_AUDITSYSCALL
2521         REG("loginuid",   S_IWUSR|S_IRUGO, loginuid),
2522         REG("sessionid",  S_IRUGO, sessionid),
2523 #endif
2524 #ifdef CONFIG_FAULT_INJECTION
2525         REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2526 #endif
2527 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2528         REG("coredump_filter", S_IRUGO|S_IWUSR, coredump_filter),
2529 #endif
2530 #ifdef CONFIG_TASK_IO_ACCOUNTING
2531         INF("io",       S_IRUGO, tgid_io_accounting),
2532 #endif
2533 };
2534
2535 static int proc_tgid_base_readdir(struct file * filp,
2536                              void * dirent, filldir_t filldir)
2537 {
2538         return proc_pident_readdir(filp,dirent,filldir,
2539                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2540 }
2541
2542 static const struct file_operations proc_tgid_base_operations = {
2543         .read           = generic_read_dir,
2544         .readdir        = proc_tgid_base_readdir,
2545 };
2546
2547 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2548         return proc_pident_lookup(dir, dentry,
2549                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2550 }
2551
2552 static const struct inode_operations proc_tgid_base_inode_operations = {
2553         .lookup         = proc_tgid_base_lookup,
2554         .getattr        = pid_getattr,
2555         .setattr        = proc_setattr,
2556 };
2557
2558 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2559 {
2560         struct dentry *dentry, *leader, *dir;
2561         char buf[PROC_NUMBUF];
2562         struct qstr name;
2563
2564         name.name = buf;
2565         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2566         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2567         if (dentry) {
2568                 if (!(current->flags & PF_EXITING))
2569                         shrink_dcache_parent(dentry);
2570                 d_drop(dentry);
2571                 dput(dentry);
2572         }
2573
2574         if (tgid == 0)
2575                 goto out;
2576
2577         name.name = buf;
2578         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2579         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2580         if (!leader)
2581                 goto out;
2582
2583         name.name = "task";
2584         name.len = strlen(name.name);
2585         dir = d_hash_and_lookup(leader, &name);
2586         if (!dir)
2587                 goto out_put_leader;
2588
2589         name.name = buf;
2590         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2591         dentry = d_hash_and_lookup(dir, &name);
2592         if (dentry) {
2593                 shrink_dcache_parent(dentry);
2594                 d_drop(dentry);
2595                 dput(dentry);
2596         }
2597
2598         dput(dir);
2599 out_put_leader:
2600         dput(leader);
2601 out:
2602         return;
2603 }
2604
2605 /**
2606  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2607  * @task: task that should be flushed.
2608  *
2609  * When flushing dentries from proc, one needs to flush them from global
2610  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2611  * in. This call is supposed to do all of this job.
2612  *
2613  * Looks in the dcache for
2614  * /proc/@pid
2615  * /proc/@tgid/task/@pid
2616  * if either directory is present flushes it and all of it'ts children
2617  * from the dcache.
2618  *
2619  * It is safe and reasonable to cache /proc entries for a task until
2620  * that task exits.  After that they just clog up the dcache with
2621  * useless entries, possibly causing useful dcache entries to be
2622  * flushed instead.  This routine is proved to flush those useless
2623  * dcache entries at process exit time.
2624  *
2625  * NOTE: This routine is just an optimization so it does not guarantee
2626  *       that no dcache entries will exist at process exit time it
2627  *       just makes it very unlikely that any will persist.
2628  */
2629
2630 void proc_flush_task(struct task_struct *task)
2631 {
2632         int i;
2633         struct pid *pid, *tgid = NULL;
2634         struct upid *upid;
2635
2636         pid = task_pid(task);
2637         if (thread_group_leader(task))
2638                 tgid = task_tgid(task);
2639
2640         for (i = 0; i <= pid->level; i++) {
2641                 upid = &pid->numbers[i];
2642                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2643                         tgid ? tgid->numbers[i].nr : 0);
2644         }
2645
2646         upid = &pid->numbers[pid->level];
2647         if (upid->nr == 1)
2648                 pid_ns_release_proc(upid->ns);
2649 }
2650
2651 static struct dentry *proc_pid_instantiate(struct inode *dir,
2652                                            struct dentry * dentry,
2653                                            struct task_struct *task, const void *ptr)
2654 {
2655         struct dentry *error = ERR_PTR(-ENOENT);
2656         struct inode *inode;
2657
2658         inode = proc_pid_make_inode(dir->i_sb, task);
2659         if (!inode)
2660                 goto out;
2661
2662         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2663         inode->i_op = &proc_tgid_base_inode_operations;
2664         inode->i_fop = &proc_tgid_base_operations;
2665         inode->i_flags|=S_IMMUTABLE;
2666
2667         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
2668                 ARRAY_SIZE(tgid_base_stuff));
2669
2670         dentry->d_op = &pid_dentry_operations;
2671
2672         d_add(dentry, inode);
2673         /* Close the race of the process dying before we return the dentry */
2674         if (pid_revalidate(dentry, NULL))
2675                 error = NULL;
2676 out:
2677         return error;
2678 }
2679
2680 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2681 {
2682         struct dentry *result = ERR_PTR(-ENOENT);
2683         struct task_struct *task;
2684         unsigned tgid;
2685         struct pid_namespace *ns;
2686
2687         result = proc_base_lookup(dir, dentry);
2688         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2689                 goto out;
2690
2691         tgid = name_to_int(dentry);
2692         if (tgid == ~0U)
2693                 goto out;
2694
2695         ns = dentry->d_sb->s_fs_info;
2696         rcu_read_lock();
2697         task = find_task_by_pid_ns(tgid, ns);
2698         if (task)
2699                 get_task_struct(task);
2700         rcu_read_unlock();
2701         if (!task)
2702                 goto out;
2703
2704         result = proc_pid_instantiate(dir, dentry, task, NULL);
2705         put_task_struct(task);
2706 out:
2707         return result;
2708 }
2709
2710 /*
2711  * Find the first task with tgid >= tgid
2712  *
2713  */
2714 struct tgid_iter {
2715         unsigned int tgid;
2716         struct task_struct *task;
2717 };
2718 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2719 {
2720         struct pid *pid;
2721
2722         if (iter.task)
2723                 put_task_struct(iter.task);
2724         rcu_read_lock();
2725 retry:
2726         iter.task = NULL;
2727         pid = find_ge_pid(iter.tgid, ns);
2728         if (pid) {
2729                 iter.tgid = pid_nr_ns(pid, ns);
2730                 iter.task = pid_task(pid, PIDTYPE_PID);
2731                 /* What we to know is if the pid we have find is the
2732                  * pid of a thread_group_leader.  Testing for task
2733                  * being a thread_group_leader is the obvious thing
2734                  * todo but there is a window when it fails, due to
2735                  * the pid transfer logic in de_thread.
2736                  *
2737                  * So we perform the straight forward test of seeing
2738                  * if the pid we have found is the pid of a thread
2739                  * group leader, and don't worry if the task we have
2740                  * found doesn't happen to be a thread group leader.
2741                  * As we don't care in the case of readdir.
2742                  */
2743                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2744                         iter.tgid += 1;
2745                         goto retry;
2746                 }
2747                 get_task_struct(iter.task);
2748         }
2749         rcu_read_unlock();
2750         return iter;
2751 }
2752
2753 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2754
2755 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2756         struct tgid_iter iter)
2757 {
2758         char name[PROC_NUMBUF];
2759         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2760         return proc_fill_cache(filp, dirent, filldir, name, len,
2761                                 proc_pid_instantiate, iter.task, NULL);
2762 }
2763
2764 /* for the /proc/ directory itself, after non-process stuff has been done */
2765 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2766 {
2767         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2768         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2769         struct tgid_iter iter;
2770         struct pid_namespace *ns;
2771
2772         if (!reaper)
2773                 goto out_no_task;
2774
2775         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2776                 const struct pid_entry *p = &proc_base_stuff[nr];
2777                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2778                         goto out;
2779         }
2780
2781         ns = filp->f_dentry->d_sb->s_fs_info;
2782         iter.task = NULL;
2783         iter.tgid = filp->f_pos - TGID_OFFSET;
2784         for (iter = next_tgid(ns, iter);
2785              iter.task;
2786              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2787                 filp->f_pos = iter.tgid + TGID_OFFSET;
2788                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2789                         put_task_struct(iter.task);
2790                         goto out;
2791                 }
2792         }
2793         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2794 out:
2795         put_task_struct(reaper);
2796 out_no_task:
2797         return 0;
2798 }
2799
2800 /*
2801  * Tasks
2802  */
2803 static const struct pid_entry tid_base_stuff[] = {
2804         DIR("fd",        S_IRUSR|S_IXUSR, fd),
2805         DIR("fdinfo",    S_IRUSR|S_IXUSR, fdinfo),
2806         REG("environ",   S_IRUSR, environ),
2807         INF("auxv",      S_IRUSR, pid_auxv),
2808         ONE("status",    S_IRUGO, pid_status),
2809         ONE("personality", S_IRUSR, pid_personality),
2810         INF("limits",    S_IRUSR, pid_limits),
2811 #ifdef CONFIG_SCHED_DEBUG
2812         REG("sched",     S_IRUGO|S_IWUSR, pid_sched),
2813 #endif
2814 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2815         INF("syscall",   S_IRUSR, pid_syscall),
2816 #endif
2817         INF("cmdline",   S_IRUGO, pid_cmdline),
2818         ONE("stat",      S_IRUGO, tid_stat),
2819         ONE("statm",     S_IRUGO, pid_statm),
2820         REG("maps",      S_IRUGO, maps),
2821 #ifdef CONFIG_NUMA
2822         REG("numa_maps", S_IRUGO, numa_maps),
2823 #endif
2824         REG("mem",       S_IRUSR|S_IWUSR, mem),
2825         LNK("cwd",       cwd),
2826         LNK("root",      root),
2827         LNK("exe",       exe),
2828         REG("mounts",    S_IRUGO, mounts),
2829         REG("mountinfo",  S_IRUGO, mountinfo),
2830 #ifdef CONFIG_PROC_PAGE_MONITOR
2831         REG("clear_refs", S_IWUSR, clear_refs),
2832         REG("smaps",     S_IRUGO, smaps),
2833         REG("pagemap",    S_IRUSR, pagemap),
2834 #endif
2835 #ifdef CONFIG_SECURITY
2836         DIR("attr",      S_IRUGO|S_IXUGO, attr_dir),
2837 #endif
2838 #ifdef CONFIG_KALLSYMS
2839         INF("wchan",     S_IRUGO, pid_wchan),
2840 #endif
2841 #ifdef CONFIG_SCHEDSTATS
2842         INF("schedstat", S_IRUGO, pid_schedstat),
2843 #endif
2844 #ifdef CONFIG_LATENCYTOP
2845         REG("latency",  S_IRUGO, lstats),
2846 #endif
2847 #ifdef CONFIG_PROC_PID_CPUSET
2848         REG("cpuset",    S_IRUGO, cpuset),
2849 #endif
2850 #ifdef CONFIG_CGROUPS
2851         REG("cgroup",  S_IRUGO, cgroup),
2852 #endif
2853         INF("oom_score", S_IRUGO, oom_score),
2854         REG("oom_adj",   S_IRUGO|S_IWUSR, oom_adjust),
2855 #ifdef CONFIG_AUDITSYSCALL
2856         REG("loginuid",  S_IWUSR|S_IRUGO, loginuid),
2857         REG("sessionid",  S_IRUSR, sessionid),
2858 #endif
2859 #ifdef CONFIG_FAULT_INJECTION
2860         REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
2861 #endif
2862 #ifdef CONFIG_TASK_IO_ACCOUNTING
2863         INF("io",       S_IRUGO, tid_io_accounting),
2864 #endif
2865 };
2866
2867 static int proc_tid_base_readdir(struct file * filp,
2868                              void * dirent, filldir_t filldir)
2869 {
2870         return proc_pident_readdir(filp,dirent,filldir,
2871                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2872 }
2873
2874 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2875         return proc_pident_lookup(dir, dentry,
2876                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2877 }
2878
2879 static const struct file_operations proc_tid_base_operations = {
2880         .read           = generic_read_dir,
2881         .readdir        = proc_tid_base_readdir,
2882 };
2883
2884 static const struct inode_operations proc_tid_base_inode_operations = {
2885         .lookup         = proc_tid_base_lookup,
2886         .getattr        = pid_getattr,
2887         .setattr        = proc_setattr,
2888 };
2889
2890 static struct dentry *proc_task_instantiate(struct inode *dir,
2891         struct dentry *dentry, struct task_struct *task, const void *ptr)
2892 {
2893         struct dentry *error = ERR_PTR(-ENOENT);
2894         struct inode *inode;
2895         inode = proc_pid_make_inode(dir->i_sb, task);
2896
2897         if (!inode)
2898                 goto out;
2899         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2900         inode->i_op = &proc_tid_base_inode_operations;
2901         inode->i_fop = &proc_tid_base_operations;
2902         inode->i_flags|=S_IMMUTABLE;
2903
2904         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
2905                 ARRAY_SIZE(tid_base_stuff));
2906
2907         dentry->d_op = &pid_dentry_operations;
2908
2909         d_add(dentry, inode);
2910         /* Close the race of the process dying before we return the dentry */
2911         if (pid_revalidate(dentry, NULL))
2912                 error = NULL;
2913 out:
2914         return error;
2915 }
2916
2917 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2918 {
2919         struct dentry *result = ERR_PTR(-ENOENT);
2920         struct task_struct *task;
2921         struct task_struct *leader = get_proc_task(dir);
2922         unsigned tid;
2923         struct pid_namespace *ns;
2924
2925         if (!leader)
2926                 goto out_no_task;
2927
2928         tid = name_to_int(dentry);
2929         if (tid == ~0U)
2930                 goto out;
2931
2932         ns = dentry->d_sb->s_fs_info;
2933         rcu_read_lock();
2934         task = find_task_by_pid_ns(tid, ns);
2935         if (task)
2936                 get_task_struct(task);
2937         rcu_read_unlock();
2938         if (!task)
2939                 goto out;
2940         if (!same_thread_group(leader, task))
2941                 goto out_drop_task;
2942
2943         result = proc_task_instantiate(dir, dentry, task, NULL);
2944 out_drop_task:
2945         put_task_struct(task);
2946 out:
2947         put_task_struct(leader);
2948 out_no_task:
2949         return result;
2950 }
2951
2952 /*
2953  * Find the first tid of a thread group to return to user space.
2954  *
2955  * Usually this is just the thread group leader, but if the users
2956  * buffer was too small or there was a seek into the middle of the
2957  * directory we have more work todo.
2958  *
2959  * In the case of a short read we start with find_task_by_pid.
2960  *
2961  * In the case of a seek we start with the leader and walk nr
2962  * threads past it.
2963  */
2964 static struct task_struct *first_tid(struct task_struct *leader,
2965                 int tid, int nr, struct pid_namespace *ns)
2966 {
2967         struct task_struct *pos;
2968
2969         rcu_read_lock();
2970         /* Attempt to start with the pid of a thread */
2971         if (tid && (nr > 0)) {
2972                 pos = find_task_by_pid_ns(tid, ns);
2973                 if (pos && (pos->group_leader == leader))
2974                         goto found;
2975         }
2976
2977         /* If nr exceeds the number of threads there is nothing todo */
2978         pos = NULL;
2979         if (nr && nr >= get_nr_threads(leader))
2980                 goto out;
2981
2982         /* If we haven't found our starting place yet start
2983          * with the leader and walk nr threads forward.
2984          */
2985         for (pos = leader; nr > 0; --nr) {
2986                 pos = next_thread(pos);
2987                 if (pos == leader) {
2988                         pos = NULL;
2989                         goto out;
2990                 }
2991         }
2992 found:
2993         get_task_struct(pos);
2994 out:
2995         rcu_read_unlock();
2996         return pos;
2997 }
2998
2999 /*
3000  * Find the next thread in the thread list.
3001  * Return NULL if there is an error or no next thread.
3002  *
3003  * The reference to the input task_struct is released.
3004  */
3005 static struct task_struct *next_tid(struct task_struct *start)
3006 {
3007         struct task_struct *pos = NULL;
3008         rcu_read_lock();
3009         if (pid_alive(start)) {
3010                 pos = next_thread(start);
3011                 if (thread_group_leader(pos))
3012                         pos = NULL;
3013                 else
3014                         get_task_struct(pos);
3015         }
3016         rcu_read_unlock();
3017         put_task_struct(start);
3018         return pos;
3019 }
3020
3021 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3022         struct task_struct *task, int tid)
3023 {
3024         char name[PROC_NUMBUF];
3025         int len = snprintf(name, sizeof(name), "%d", tid);
3026         return proc_fill_cache(filp, dirent, filldir, name, len,
3027                                 proc_task_instantiate, task, NULL);
3028 }
3029
3030 /* for the /proc/TGID/task/ directories */
3031 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3032 {
3033         struct dentry *dentry = filp->f_path.dentry;
3034         struct inode *inode = dentry->d_inode;
3035         struct task_struct *leader = NULL;
3036         struct task_struct *task;
3037         int retval = -ENOENT;
3038         ino_t ino;
3039         int tid;
3040         unsigned long pos = filp->f_pos;  /* avoiding "long long" filp->f_pos */
3041         struct pid_namespace *ns;
3042
3043         task = get_proc_task(inode);
3044         if (!task)
3045                 goto out_no_task;
3046         rcu_read_lock();
3047         if (pid_alive(task)) {
3048                 leader = task->group_leader;
3049                 get_task_struct(leader);
3050         }
3051         rcu_read_unlock();
3052         put_task_struct(task);
3053         if (!leader)
3054                 goto out_no_task;
3055         retval = 0;
3056
3057         switch (pos) {
3058         case 0:
3059                 ino = inode->i_ino;
3060                 if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0)
3061                         goto out;
3062                 pos++;
3063                 /* fall through */
3064         case 1:
3065                 ino = parent_ino(dentry);
3066                 if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0)
3067                         goto out;
3068                 pos++;
3069                 /* fall through */
3070         }
3071
3072         /* f_version caches the tgid value that the last readdir call couldn't
3073          * return. lseek aka telldir automagically resets f_version to 0.
3074          */
3075         ns = filp->f_dentry->d_sb->s_fs_info;
3076         tid = (int)filp->f_version;
3077         filp->f_version = 0;
3078         for (task = first_tid(leader, tid, pos - 2, ns);
3079              task;
3080              task = next_tid(task), pos++) {
3081                 tid = task_pid_nr_ns(task, ns);
3082                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3083                         /* returning this tgid failed, save it as the first
3084                          * pid for the next readir call */
3085                         filp->f_version = (u64)tid;
3086                         put_task_struct(task);
3087                         break;
3088                 }
3089         }
3090 out:
3091         filp->f_pos = pos;
3092         put_task_struct(leader);
3093 out_no_task:
3094         return retval;
3095 }
3096
3097 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3098 {
3099         struct inode *inode = dentry->d_inode;
3100         struct task_struct *p = get_proc_task(inode);
3101         generic_fillattr(inode, stat);
3102
3103         if (p) {
3104                 stat->nlink += get_nr_threads(p);
3105                 put_task_struct(p);
3106         }
3107
3108         return 0;
3109 }
3110
3111 static const struct inode_operations proc_task_inode_operations = {
3112         .lookup         = proc_task_lookup,
3113         .getattr        = proc_task_getattr,
3114         .setattr        = proc_setattr,
3115 };
3116
3117 static const struct file_operations proc_task_operations = {
3118         .read           = generic_read_dir,
3119         .readdir        = proc_task_readdir,
3120 };