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