git.oblomov.eu Git - linux-2.6/blob - fs/proc/base.c

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